perf: allow BATCH_PULL_SIZE in any mode for benchmarking

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

.agents/skills/adding-a-trigger/SKILL.md

@@ -1,282 +0,0 @@
----
-name: adding-a-trigger
-description: Checklist for adding a new TriggerCrud-based trigger type to Windmill (Azure, GCP, Kafka, etc.). Use when wiring a new trigger kind across backend, frontend, CLI, and capture infrastructure.
----
-
-# Skill: Adding a New Trigger Type
-
-Use this skill when adding a trigger kind that implements `TriggerCrud` (Kafka, GCP, Azure, MQTT, SQS, NATS, Postgres, Email…). For native triggers (Nextcloud, Google Drive — things wired through `windmill-native-triggers`), use the `native-trigger` skill instead.
-
-The goal of this doc is to enumerate every file that needs to change. Missing any one of them leads to silent regressions: sync drops the trigger, capture button does nothing, workspace forks lose it, sidebar counters undercount. Follow the checklist top-to-bottom — each section is independent enough to be validated on its own.
-
-Throughout this doc, substitute `{kind}` for the new trigger kind (`azure`, `kafka`, …), `{Kind}` for PascalCase (`Azure`, `Kafka`), `{KIND}` for SCREAMING (`AZURE`, `KAFKA`).
-
-## Reference implementations
-
-- **GCP** — closest analogue to Azure. Has push + pull, OIDC auth, ARM-like resource paths, capture handler. Grep for `gcp_trigger` / `GcpTrigger`.
-- **Kafka** — simpler (pull-only, streaming). Good for trivial integrations.
-- **Azure** — most recently added (2026). Shared-secret push auth, Event Grid namespaces + basic topics, ARM resource discovery, Namespace-pull data-plane. Grep for `azure_trigger` / `AzureTrigger`.
-
-## 1. Database migration
-
-Create a migration: `cargo sqlx migrate add -r add_{kind}_trigger` from `backend/`. Never write timestamps manually.
-
-The `up.sql` usually defines:
-- An optional enum type (e.g. `AZURE_MODE`) if the trigger has sub-kinds
-- The `{kind}_trigger` table with at minimum these columns (mirrored from kafka/gcp):
-  - primary: `(workspace_id, path)`
-  - `script_path`, `is_flow`, `enabled`, `mode`, `permissioned_as`, `edited_by`, `email`
-  - `edited_at`, `error`, `server_id`, `last_server_ping`
-  - `error_handler_path`, `error_handler_args jsonb`, `retry jsonb`
-  - trigger-specific fields
-- Indexes on foreign keys + any frequently-filtered columns
-- Foreign key to `workspace`
-- The RLS policies (`see_own`, `see_member`, `see_folder_extra_perms_user_*`, `see_extra_perms_user_*`, `see_extra_perms_groups_*`), copied from an existing trigger table
-
-**RLS: wrap every session GUC read in a scalar sub-select.** Write the session
-reads as `(select current_setting('session.user'))`,
-`= any((select regexp_split_to_array(current_setting('session.groups'), ','))::text[])`,
-`?| (select regexp_split_to_array(current_setting('session.pgroups'), ','))::text[]`,
-`? (select concat('u/', current_setting('session.user')))`, etc. — not the bare
-`current_setting(...)`. The GUCs are set with `SET LOCAL`, so the sub-select
-hoists them to a one-time InitPlan instead of re-evaluating per scanned row.
-Put the `::text[]` cast **outside** the sub-select for the array cases: in an
-`= any (...)` context, casting inside — `= any((select ...::text[]))` — makes
-Postgres parse the operand as a row-returning subquery and fails at CREATE with
-`operator does not exist: text = text[]`. The outside cast keeps it in
-array-operand form. See migration `20260714230440_wrap_session_gucs_in_rls_policies`
-for the canonical wrapped forms.
-
-Down migration drops the table and any enum types.
-
-## 2. Backend crate (`windmill-trigger-{kind}`)
-
-Create a new crate under `backend/windmill-trigger-{kind}/` with:
-
-- `Cargo.toml`: features `enterprise`, `private` if EE, standard deps
-- `src/lib.rs`: `pub use mod_ee::*;` behind `#[cfg(all(feature = "enterprise", feature = "private"))]`
-- `src/mod_ee.rs`: core types + helpers
-- `src/handler_ee.rs`: `TriggerCrud` impl + route handlers
-- `src/listener_ee.rs`: (only if streaming/pull-based) `Listener` trait impl
-
-Required in `mod_ee.rs`:
-- `{Kind}Config` struct (persisted shape, `FromRow`)
-- `{Kind}ConfigRequest` struct (what API receives — usually similar to Config but with validation fields)
-- `{Kind}Trigger` unit struct (implements the traits)
-- `impl TriggerJobArgs for {Kind}Trigger` — sets `TRIGGER_KIND`, `Payload`, `v1_payload_fn`
-
-Required in `handler_ee.rs`:
-- `#[async_trait] impl TriggerCrud for {Kind}Trigger` with:
-  - `type Trigger = Trigger<{Kind}Config>`
-  - `type TriggerConfigRequest = {Kind}ConfigRequest`
-  - `const ROUTE_PREFIX: &'static str = "/{kind}_triggers";`
-  - `const TABLE_NAME`, `ADDITIONAL_SELECT_FIELDS`
-  - `get_deployed_object`, `validate_config`, `create_trigger`, `update_trigger`, `delete_trigger`, `test_connection`
-  - `additional_routes` (optional — mount extra endpoints for things like ARM resource listing, topic discovery)
-
-Register the crate in `backend/Cargo.toml` as a workspace member and as a dep of `windmill-api` behind the feature flag.
-
-## 3. Wire into `windmill-api` (feature-gated everywhere)
-
-**`backend/windmill-api/src/triggers/handler.rs`** — mount the trigger crate:
-```rust
-#[cfg(all(feature = "enterprise", feature = "{kind}_trigger", feature = "private"))]
-{
-    use crate::triggers::{kind}::{Kind}Trigger;
-    router = router.nest({Kind}Trigger::ROUTE_PREFIX, complete_trigger_routes({Kind}Trigger));
-}
-```
-
-**`backend/windmill-api/src/triggers/{kind}/mod.rs`** — re-export the crate:
-```rust
-pub use windmill_trigger_{kind}::*;
-```
-
-**`backend/windmill-api/src/lib.rs`** — if the trigger receives inbound pushes, add a webhook route:
-```rust
-.nest("/{kind}/w/{workspace_id}", {
-    #[cfg(all(feature = "enterprise", feature = "{kind}_trigger", feature = "private"))]
-    { triggers::{kind}::handler_oss::{kind}_push_route_handler() }
-    #[cfg(not(...))]
-    { Router::new() }
-})
-```
-
-## 4. `TriggerKind` enum (`backend/windmill-types/src/triggers.rs`)
-
-Already has slots for most triggers but verify your variant exists:
-- Add `{Kind}` to the `TriggerKind` enum
-- Add match arm in `to_key()`
-- Add match arm in `from_str`
-- Add match arm in `JobTriggerKind` (if jobs need kind tagging)
-
-## 5. OpenAPI (`backend/windmill-api/openapi.yaml`)
-
-This file is huge and the single most-forgotten place. Add:
-
-- `/w/{workspace}/{kind}_triggers/create` + `/update/{path}` + `/delete/{path}` + `/get/{path}` + `/list` + `/exists/{path}` + `/setmode/{path}` + `/test` paths (mirror gcp section)
-- Any `additional_routes` your handler exposes (resource discovery, etc.)
-- Schemas: `{Kind}Trigger`, `{Kind}TriggerData`, `{Kind}Mode` (if enum), `{Kind}DeliveryConfig`, helper request/response types
-- Add `{kind}` to `CaptureTriggerKind` enum
-- Add `{kind}_used: boolean` to the `UsedTriggers` response schema
-
-Regenerate frontend client: `npm run generate-backend-client` from `frontend/`.
-
-## 6. `UsedTriggers` + workspace export
-
-**`backend/windmill-api-workspaces/src/workspaces.rs`** — add `{kind}_used: bool` to the `UsedTriggers` struct and add an `EXISTS(SELECT 1 FROM {kind}_trigger …)` to the `get_used_triggers` query.
-
-**`backend/windmill-api/src/workspaces_export.rs`** — add export block mirroring gcp's (export lists all triggers, serializes them to YAML/JSON). The block re-uses the `trigger_ignore_keys` variable so the new kind automatically participates in fork-export stripping (`mode` field is omitted when the source workspace is a fork — keeps fork→parent merges from flipping the parent's enabled state).
-
-**Fork cloning (`clone_triggers_and_schedules` in workspaces.rs)** — add an `INSERT INTO {kind}_trigger ... SELECT ...` block that copies all rows from the parent workspace, forcing `mode = 'disabled'::TRIGGER_MODE`. Always runs at fork creation; forgetting this means users can't carry `{kind}` triggers into their forks.
-
-## 6.5 Hardcoded trigger-kind arrays (silent-failure hotspots)
-
-Several files keep **hardcoded arrays** of trigger kind strings. Miss one and ACL checks / user offboarding / trash drop your kind:
-
-- **`backend/windmill-api-groups/src/granular_acls.rs`** — `KINDS: [&str; N]`. **Increment N** (the compile error is cryptic otherwise). Controls which kinds accept granular ACL operations.
-- **`backend/windmill-api-users/src/users.rs`** (`extra_perms_tables`) — which tables get `extra_perms` entries cleaned when a user is deleted.
-- **`backend/windmill-api/src/offboarding.rs`** — three separate arrays (enumeration, fork-copy, and delete paths). **All three** need the new kind.
-- **`backend/windmill-api/src/trash.rs`** — `valid_tables` for the trash / restore API.
-- **`backend/windmill-git-sync/src/lib.rs`** — add a test assertion for `DeployedObject::{Kind}Trigger.get_kind() == "{kind}_trigger"` (the `get_kind` match arm itself lives in the enum impl — already required by the Rust compiler).
-- **`backend/windmill-api-auth/src/scopes.rs`** — add the `{Kind}Triggers` variant to `ScopeDomain` enum + `as_str` match + `from_str` match. Required for the OAuth/token system to recognise `{kind}_triggers:read|write` scopes.
-- **`backend/windmill-api/src/token.rs`** (`build_trigger_scope_domains` → `TRIGGER_DOMAINS`) — add `("{kind}_triggers", "{Kind display name}")` so the CreateToken UI's scope selector surfaces the `read` / `write` checkboxes.
-
-**OpenAPI enums** to extend (do NOT forget — generated client will allow it but server rejects as 400):
-- `CaptureTriggerKind` enum
-- Three `kind` enums under `/w/{workspace}/acls/{get,add,remove}/{kind}/{path}` (yes, same list repeated three times)
-
-After editing any of these, run a full `cargo check` with your feature flag + `gcp_trigger` + other core flags — the `KINDS: [&str; N]` length mismatch only surfaces when the crate compiles.
-
-## 7. Capture infrastructure (`backend/windmill-api/src/capture.rs`)
-
-If the trigger supports push delivery, it also needs a capture endpoint so users can test it:
-
-- `{Kind}TriggerConfig` struct (gated by feature flags)
-- `TriggerConfig::{Kind}` variant
-- `set_{kind}_trigger_config` function (creates the subscription/equivalent pointing at the capture URL — use your `manage_{kind}_subscription` helper with `trigger_mode=false`)
-- Both real + no-op versions behind feature gates
-- `TriggerKind::{Kind} => set_{kind}_trigger_config(...)` arm in `set_config`
-- `{kind}_payload` async handler — validates auth (if any), processes payload, calls `insert_capture_payload`
-- Route: `.route("/{kind}/{runnable_kind}/{*path}", post({kind}_payload))` inside `workspaced_unauthed_service` — and expand the surrounding `#[cfg(any(...))]` to include your feature flag
-
-## 8. CLI (`cli/`) — easy to miss, breaks sync silently
-
-Check all of these:
-
-**`cli/src/types.ts`:**
-- Add `"{kind}"` to `TRIGGER_TYPES` array
-- Add `"{kind}_trigger"` to `getTypeStrFromPath` return union
-- Add match case in `getTypeStrFromPath`'s `typeEnding ===` chain
-- Add `pushTrigger("{kind}", ...)` branch in `pushObj`
-
-**`cli/src/commands/trigger/trigger.ts`:**
-- Import `{Kind}Trigger` type
-- Add `{kind}: {Kind}Trigger` to the `Trigger` type map
-- Add `{kind}: wmill.get{Kind}Trigger`, `update{Kind}Trigger`, `create{Kind}Trigger` to each function map
-- Add `{kind}: { ... }` template to `triggerTemplates`
-- Add `list{Kind}Triggers` call + spread in the `list` aggregation
-- Update `--kind` option descriptions to mention the new kind
-
-**`cli/src/commands/sync/sync.ts`:**
-- Add `path.endsWith(".{kind}_trigger" + ext)` in the file-type filter
-- Add `typ == "{kind}_trigger"` in `getTypeOrder`
-- Add `"{kind}_trigger"` to the delete-suffix regex (~line 3092)
-- Add a `case "{kind}_trigger"` in the delete switch
-
-**`cli/src/guidance/skills.ts`** — **DO NOT EDIT DIRECTLY**. It's auto-generated by `system_prompts/generate.py`. Instead:
-- Edit `system_prompts/utils.py` → append `('{Kind}Trigger', '{kind}_trigger')` to the `SCHEMA_MAPPINGS['triggers']` list (this is the master list — the one in `generate.py` is duplicated and `utils.py` wins)
-- Then run `python3 system_prompts/generate.py` — it regenerates `cli/src/guidance/skills.ts` with the schema extracted from `backend/windmill-api/openapi.yaml`
-- Commit the regenerated file
-
-## 9. Frontend — editor + drawer
-
-Under `frontend/src/lib/components/triggers/{kind}/`:
-
-- `{Kind}TriggerPanel.svelte` — the tile shown in the triggers listing
-- `{Kind}TriggerEditor.svelte` — outer drawer wrapper
-- `{Kind}TriggerEditorInner.svelte` — state + business logic; must expose:
-  - `openEdit(path, isFlow, defaultValues?)` method
-  - `isEditor` prop, `onConfigChange` + `onCaptureConfigChange` callbacks
-  - `get{Kind}Config()` + `get{Kind}CaptureConfig()` helpers
-  - `captureConfig = $derived.by(untrack(() => isEditor) ? get{Kind}CaptureConfig : () => ({}))`
-  - `$effect(() => { const args = [captureConfig, isValid] as const; untrack(() => onCaptureConfigChange?.(...args)) })`
-- `{Kind}TriggerEditorConfigSection.svelte` — form fields; use design-system components (`TextInput`, `Select`, `Toggle`, `ToggleButtonGroup`), never raw `<input>`
-- `{Kind}Capture.svelte` — capture panel; wraps `CaptureSection` with `captureType="{kind}"`
-- `utils.ts` — `requestBody` builders and any trigger-type-specific helpers
-
-## 10. Frontend — global integration
-
-Easy to miss:
-
-- **`frontend/src/lib/components/triggers.ts`** — add `'{kind}'` to the `TriggerKind` union
-- **`frontend/src/lib/components/triggers/CaptureWrapper.svelte`**:
-  - Import `{Kind}Capture`
-  - Add to `isStreamingCapture()` array (streaming = pull-style; push-style is typically `false`)
-  - Add `{:else if captureType === '{kind}'}` branch with the `<{Kind}Capture>` render
-- **`frontend/src/lib/components/sidebar/SidebarContent.svelte`** — import the icon, add the nav entry
-- **`frontend/src/lib/components/sidebar/OperatorMenu.svelte`** — add the operator-mode entry
-- **`frontend/src/routes/(root)/(logged)/+layout.svelte`** — destructure `{kind}_used` from `/get_used_triggers` response, push `'{kind}'` into `usedKinds`
-- **`frontend/src/lib/components/search/GlobalSearchModal.svelte`** — import icon, add "Go to {Kind} ..." entry
-- **`frontend/src/lib/components/offboarding-utils.ts`** — add mappings `{kind}_trigger: '{kind}_triggers'` and `{kind}_trigger: '{kind} trigger'`
-- **`frontend/src/lib/components/icons/{Kind}Icon.svelte`** — single-path SVG, `fill={color ?? 'currentColor'}`, `size` prop default 16 (match existing icons — don't hardcode colors, don't use `width`/`height` props)
-- **`frontend/src/routes/(root)/(logged)/{kind}_triggers/+page.svelte`** — listing page (mirror `gcp_triggers/+page.svelte` for push+pull, `kafka_triggers` for pure streaming)
-- **`frontend/src/lib/components/CompareWorkspaces.svelte`** — workspace fork / compare tool. Needs: service import, editor import, `{kind}Editor` `$state`, `case '{kind}'` in `openTriggerDetails()`, entry in `triggerServices` object (list/delete/normalize), and `<{Kind}TriggerEditor bind:this={{kind}Editor} />` in the template
-
-## 10.5 AI system prompts (`system_prompts/`)
-
-- **`system_prompts/utils.py`** — append `('{Kind}Trigger', '{kind}_trigger')` to `SCHEMA_MAPPINGS['triggers']` (master list used by code generation + CLI skills)
-- **`system_prompts/generate.py`** — also has a duplicated `schema_types` list (~line 903) for the AI `triggers` skill content. Add `('{Kind}Trigger', '{kind}_trigger')` there too
-- **`system_prompts/generate.py`** `schema_names` (~line 1192) — add `'{Kind}Trigger'` (add `'New{Kind}Trigger'` only if the OpenAPI declares one; GCP and Azure don't)
-- Run `python3 system_prompts/generate.py` — this rewrites `cli/src/guidance/skills.ts` and all `auto-generated/` docs. Commit the regenerated files
-
-## 11. Validation
-
-Run all of these before declaring done:
-
-```bash
-# Backend
-cd backend
-cargo check --features enterprise,{kind}_trigger,private   # minimal
-cargo check --features enterprise,azure_trigger,private,gcp_trigger,http_trigger,mqtt_trigger,postgres_trigger,sqs_trigger,kafka,nats,smtp,websocket  # full
-
-# SQLx offline data (never run `cargo sqlx prepare` directly — use the wrapper)
-./update_sqlx.sh
-
-# Frontend
-cd frontend
-npm run generate-backend-client
-npm run check:fast
-```
-
-Smoke test in the UI: create a trigger, save, check it appears in sidebar + search, delete, re-create via CLI `wmill sync`.
-
-## 12. Common pitfalls
-
-- **Forgetting feature gates in `workspaced_unauthed_service()`** — the surrounding `#[cfg(any(...))]` expression must include your feature flag, not just the inner `#[cfg]` on the route
-- **`.route(path, ...).route(path, ...)` with same path and different methods** — older axum replaced; use `.route(path, post(h1).options(h2))` to chain methods on the same `MethodRouter`
-- **`on:event` directives** — legacy Svelte 4, no-op in runes mode. Use callback props (`onSelected`, `onConfigChange`)
-- **`$bindable(default_value)` on optional props** — banned by project CLAUDE.md. Use `$bindable()` + `$derived(prop ?? default)` instead
-- **CORS layer intercepting OPTIONS** — tower-http CorsLayer short-circuits OPTIONS before reaching your handler. For server-to-server webhook endpoints, drop the CORS layer entirely (CORS is browser-only)
-- **DeliveryAttributeMappings / custom headers for auth** — prefer HMAC or sha256-hashed shared secrets over opaque JWTs when the provider doesn't support signed tokens natively. Store only the hash; regenerate secret on every save
-- **ARM / API resource-listing cascades** — if the trigger's resource type is deep (Azure: subscription → RG → namespace → topic), offer dropdowns in the UI populated from the provider's APIs using the user's credential resource
-- **Clearing stale selections on dependency change** — when a dropdown's underlying data reloads (e.g., user changes SP or edition), clear selections that no longer match the new list
-- **Workspace-scoped tag compatibility** — if the trigger has tags, verify forked workspaces handle them (see commit `0773b5bc85` for a historical fix)
-
-## 13. EE file split
-
-If the trigger is enterprise-only, the code lives in `windmill-ee-private__worktrees/.../windmill-trigger-{kind}/src/*_ee.rs` and is symlinked into the OSS tree. The `windmill-ee-private__worktrees/` directory holds the real files; changes propagate via symlinks. See `docs/enterprise.md` for the workflow.
-
-## 14. Final checklist before PR
-
-- [ ] Migration up/down tested (revert + re-apply)
-- [ ] `./update_sqlx.sh` committed the updated `.sqlx/` offline data
-- [ ] `cargo check` passes with your feature flag + with all trigger features
-- [ ] `npm run check:fast` passes
-- [ ] Trigger visible in sidebar with correct icon weight (not oversized/colored — use `currentColor`)
-- [ ] Create, edit, delete flow all work in the UI
-- [ ] Capture button works (if push-capable)
-- [ ] Trigger appears in `/get_used_triggers` → sidebar pulse
-- [ ] `wmill sync pull` + `wmill sync push` both round-trip the trigger
-- [ ] `wmill trigger list` includes it
-- [ ] OpenAPI schemas are complete (no `null` in generated types)

.agents/skills/ai-chat/SKILL.md

@@ -1,40 +0,0 @@
----
-name: ai-chat
-description: Guidance for improving the Windmill AI chat (copilot), especially global mode — tools, prompts, and context-window discipline. Use when editing chat tools, system prompts, or tool-result shapes under frontend/src/lib/components/copilot/chat, or when changing how the chat manages its context window.
----
-
-## Always benchmark before and after
-
-No context or behavior change ships without an `ai_evals` A/B on the affected mode.
-Add or adjust cases for exactly what you changed — see the `ai-evals` skill for
-authoring and the full run reference.
-
-Run the affected mode **before** your change and **after**, same model(s), same cases.
-
-## Measure the window first, and cumulative second
-
-Optimize **`finalContextTokens`** (window occupancy — what drives overflow and
-compaction), then cumulative prompt tokens.
-
-## Context discipline
-
-The dominant fixed cost is per-iteration overhead: the system prompt **plus every
-tool schema** is re-sent on every loop iteration. So:
-
-- **Every tool and every parameter is a permanent tax.** Justify each one and measure
-  it; an extra "locate" round-trip can cost more than the reads it saves. Strip dead
-  params rather than leaving them in the schema.
-- **Tool results return the minimum.** Never echo content the model already has. The
-  canonical mistake: a write tool that returns the whole edited artifact right after
-  the model authored it — return `{ success, message }` instead. When you touch a
-  *shared* write helper (e.g. `finishAppDraftWrite` in `global/core.ts`), re-check
-  this invariant for **all** the write tools routing through it — the echo has
-  regressed before via a shared refactor.
-
-## Prompts and tool descriptions are part of the surface
-
-The system prompt and tool descriptions steer behavior as much as the tools
-themselves, and are benchmarkable the same way. A description that advertises
-truncation makes the model self-limit; the path-conventions block changes where
-drafts land. Treat prompt/description edits as real changes and A/B them — a
-pure-prompt change is a legitimate, measurable improvement.

.agents/skills/ai-evals/SKILL.md

@@ -1,87 +0,0 @@
----
-name: ai-evals
-description: Author and run black-box benchmark cases for the Windmill AI generation modes (flow/app/script/cli/global) in ai_evals/. Use when adding or changing eval cases, or when running before/after benchmarks for AI chat / copilot changes.
----
-
-# AI evals — authoring and running benchmark cases
-
-`ai_evals/` is a black-box benchmark runner for the Windmill AI generation modes:
-`flow`, `app`, `script`, `cli`, `global`. It always tests the **current** production
-prompts, tools, and guidance in this checkout. Each attempt runs the real production
-path, deterministic validation, then LLM judging.
-
-The goal is to test current production guidance with realistic user requests — **not**
-to pin one exact implementation shape.
-
-## Running benchmarks
-
-```bash
-cd ai_evals
-bun install                       # first time; frontend modes also need `cd frontend && bun install`
-bun run cli -- models             # list model aliases
-bun run cli -- cases global       # list cases for a mode
-bun run cli -- run global global-test1-script-create --model sonnet
-```
-
-Frontend modes (`flow`/`script`/`app`/`global`) route model calls through a Windmill
-backend's `/api/w/<ws>/ai/proxy`, so you need **any** reachable backend:
-
-```bash
-WMILL_AI_EVAL_BACKEND_URL=http://127.0.0.1:<port> WMILL_AI_EVAL_BACKEND_WORKSPACE=integration-tests \
-  bun run cli -- run global <caseIds...> --models sonnet,gpt-5.5,gemini-3.1-pro-preview
-```
-
-- **Reuse an existing workspace.** CE builds cap workspaces, so temp-workspace
-  creation 400s ("reached workspace limit"). Always set
-  `WMILL_AI_EVAL_BACKEND_WORKSPACE=integration-tests` (or any existing workspace) to
-  reuse one. The only side effect of a run is upserting an `f/evals/ai/<provider>`
-  resource there.
-- Provider keys live in `ai_evals/.env` and are auto-loaded by bun. The judge is a
-  separate Anthropic call (default `claude-sonnet-4-6`) regardless of the model under
-  test.
-
-## Authoring core rules
-
-1. Write prompts like a real user request.
-2. Prefer behavior, inputs, constraints, and outcomes over internal implementation.
-3. Keep deterministic validation narrow and hard.
-4. Put semantic expectations in `judgeChecklist`.
-5. Use `expected` fixtures only when exact structure really matters.
-
-### Prompt writing
-
-Prompts should sound like something a user would naturally ask. Do not write prompts
-as if the user knows Windmill internals unless the case explicitly tests a power-user
-workflow.
-
-Good:
-- "Create a flow that routes support requests based on customer tier."
-- "Add a reset button that sets the counter back to 0."
-- "Create a flow that reuses the existing greeting script instead of duplicating the logic."
-
-Bad:
-- "Use `branchone` with 3 branches and a default branch."
-- "Create a `rawscript` step with this exact topology."
-- "This is a benchmark harness."
-
-### Deterministic validation
-
-Use deterministic checks only for hard failures: missing required files; unexpected
-extra files when the prompt says not to create them; syntax errors; unresolved flow
-refs; missing required special modules or suspend config; obvious corruption.
-
-Do **not** encode one preferred implementation. Bad hard checks: exact step topology
-for a creation flow; exact branch structure when the prompt only asked for routing;
-exact input shape when multiple reasonable shapes are acceptable.
-
-### Judge checklist
-
-Every non-trivial case should have a `judgeChecklist` capturing user-visible behavior
-that must be present, important constraints, and key completion criteria — not
-low-level implementation details unless truly required.
-
-Good: "the flow calculates the order total with 8% tax"; "the flow reuses the existing
-workspace script instead of rewriting the logic". Bad: "uses `branchone`"; "contains a
-`rawscript` node".
-
-See `ai_evals/README.md` for the full case format, fields, and fixture details.

.agents/skills/commit/SKILL.md

@@ -1,6 +1,5 @@
 ---
 name: commit
-user_invocable: true
 description: Create a git commit with conventional commit format. MUST use anytime you want to commit changes.
 ---
 
@@ -53,6 +52,8 @@ chore: upgrade sqlx to 0.7
 4. Stage ONLY the modified/relevant files: `git add <file1> <file2> ...`
 5. Create the commit with conventional format:
    ```bash
-   git commit -m "<type>: <description>"
+   git commit -m "<type>: <description>
+
+   Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>"
    ```
 6. Run `git status` to verify the commit succeeded

.agents/skills/local-review-codex/SKILL.md

@@ -1,50 +0,0 @@
----
-name: local-review-codex
-description: Run the CI Codex PR review locally against this branch's unpushed work (committed + uncommitted) before pushing. Same policy, model, and reasoning effort as the codex-pr-review GitHub action.
----
-
-# Local Codex Review (pre-push)
-
-Runs the exact same review Codex performs in CI (`.github/workflows/codex-pr-review.yml`),
-but locally and scoped to work you have not pushed yet — so you catch what CI would flag
-before the PR exists. Use this before `git push` on a non-trivial change.
-
-**Correspondence with CI** — identical:
-- Policy: `REVIEW.md` (severity triage, public-surface checklist, AGENTS.md compliance, test coverage).
-- Model: `gpt-5.6-sol`, `model_reasoning_effort="xhigh"`.
-- Output: markdown starting with `## Codex Review`, findings tagged P0 / P1 / P2 with file:line.
-
-**Differences from CI** — local-only:
-- Scope is the current branch vs `main` at the merge-base, **including uncommitted changes** (CI reviews a pushed PR diff).
-- Sandbox is `read-only` (CI uses `danger-full-access` on an ephemeral runner). Codex reads the diff and files but cannot modify your working tree.
-- Fresh context is inherent: `codex exec` is a separate cold process, so it does not anchor on the current chat session — the same reason `local-review` insists on a subagent.
-
-## Prerequisites
-
-- `codex` CLI **>= 0.144.1** installed and authed (`codex login` or `OPENAI_API_KEY`). Older CLIs reject `gpt-5.6-sol` with "requires a newer version of Codex". Upgrade with `npm install --global @openai/codex@0.144.1` (may need `sudo` for a global install). Keep this in sync with the pin in `.github/workflows/codex-pr-review.yml`.
-- `git fetch` the base ref if it's stale, so the merge-base is accurate.
-
-## Run
-
-```bash
-bash .agents/skills/local-review-codex/run.sh          # review vs main (default)
-bash .agents/skills/local-review-codex/run.sh <base>   # review vs a different base ref
-```
-
-Invoke with `bash` (or run the executable directly) — the script needs Bash for
-`set -o pipefail`; `sh` is Dash on Debian/Ubuntu and would fail. If `main` isn't a
-local branch (e.g. a fresh single-branch checkout), the runner falls back to
-`origin/main` automatically.
-
-The script computes `BASE_SHA = git merge-base HEAD <base>`, feeds Codex `REVIEW.md` plus a
-diff context pointing at `git diff <BASE_SHA>` (which folds in uncommitted edits), and prints
-the review. It writes only temp files — nothing lands in the working tree.
-
-## Relaying the result
-
-Print the Codex output verbatim. Do not re-summarize or filter it — the value of a cold Codex
-pass is surfacing what the current session would rationalize away. Then decide with the user
-whether to address findings before pushing.
-
-For a Claude-native review instead, use `local-review` (branch-diff-reviewer subagent). This
-skill is the Codex counterpart; run both for independent perspectives.

.agents/skills/local-review-codex/run.sh

@@ -1,91 +0,0 @@
-#!/usr/bin/env bash
-# Local Codex review — mirrors the .github/workflows/codex-pr-review.yml CI job,
-# but scoped to this branch's unpushed work (committed + uncommitted) so you can
-# review before pushing. Same policy (REVIEW.md), same model (gpt-5.6-sol) and
-# reasoning effort (xhigh) as CI. Runs read-only: Codex cannot modify your tree.
-#
-# Usage: run.sh [BASE_REF]   (BASE_REF defaults to "main")
-set -euo pipefail
-
-BASE_REF="${1:-main}"
-REPO_ROOT="$(git rev-parse --show-toplevel)"
-cd "$REPO_ROOT"
-
-if ! command -v codex >/dev/null 2>&1; then
-  echo "codex CLI not found. Install with: npm install --global @openai/codex@0.144.1" >&2
-  exit 1
-fi
-
-# Resolve the base to a concrete commit, preferring a local ref but falling back to
-# the remote-tracking ref — checkouts (CI, single-branch clones) often have only
-# origin/main, not a local main.
-if git rev-parse --verify --quiet "${BASE_REF}^{commit}" >/dev/null; then
-  BASE_COMMITISH="$BASE_REF"
-elif git rev-parse --verify --quiet "origin/${BASE_REF}^{commit}" >/dev/null; then
-  BASE_COMMITISH="origin/${BASE_REF}"
-else
-  echo "Base ref '$BASE_REF' not found as '$BASE_REF' or 'origin/$BASE_REF'. Try: git fetch origin $BASE_REF" >&2
-  exit 1
-fi
-
-# Diff from the merge-base so only this branch's changes are reviewed. Using the
-# base SHA with a single-ref `git diff` also folds in uncommitted working-tree edits,
-# but `git diff` never sees untracked files — those are gathered separately below so
-# brand-new files (a whole new module, a new skill dir) are not silently skipped.
-BASE_SHA="$(git merge-base HEAD "$BASE_COMMITISH")"
-HEAD_SHA="$(git rev-parse HEAD)"
-UNTRACKED="$(git ls-files --others --exclude-standard)"
-
-if [ "$BASE_SHA" = "$HEAD_SHA" ] && git diff --quiet "$BASE_SHA" && [ -z "$UNTRACKED" ]; then
-  echo "No changes vs $BASE_REF — nothing to review." >&2
-  exit 0
-fi
-
-PROMPT="$(mktemp)"
-OUT="$(mktemp)"
-trap 'rm -f "$PROMPT" "$OUT"' EXIT
-
-# REVIEW.md is the shared policy CI feeds Codex. Append the local output-format
-# and diff context inline (CI reads these from a generated context file; inlining
-# keeps the working tree clean — no scratch files land in the repo).
-cat REVIEW.md > "$PROMPT"
-cat >> "$PROMPT" <<EOF
-
-# Codex output format
-
-- This is a pre-push LOCAL review of unpushed work; there is no PR yet.
-- Inspect the changes by running the diff commands in the review context below.
-- Untracked files do NOT appear in \`git diff\`. Review every untracked path listed below by reading it directly (\`cat\`) — treat its entire contents as newly added.
-- Return markdown starting with \`## Codex Review\`.
-- Tag each finding with a severity (P0 / P1 / P2), file path, and line number when known confidently.
-
-# Review context
-
-Local review (pre-push): current branch vs $BASE_REF
-Base SHA: $BASE_SHA
-Head SHA: $HEAD_SHA (plus any uncommitted working-tree changes)
-
-Changed commits command:
-git log --oneline $BASE_SHA..HEAD
-
-Changed files command:
-git diff --stat $BASE_SHA
-
-Full review diff command (tracked changes, includes uncommitted edits):
-git diff --unified=0 $BASE_SHA
-
-Untracked files (NOT in the diff above — read each one directly, it is entirely new):
-$(if [ -n "$UNTRACKED" ]; then printf '%s\n' "$UNTRACKED"; else echo "(none)"; fi)
-EOF
-
-codex exec \
-  -C "$REPO_ROOT" \
-  -m gpt-5.6-sol \
-  -c 'model_reasoning_effort="xhigh"' \
-  -s read-only \
-  -o "$OUT" \
-  - < "$PROMPT"
-
-echo
-echo "===== Codex review ====="
-cat "$OUT"

.agents/skills/local-review/SKILL.md

@@ -1,98 +1,97 @@
 ---
 name: local-review
-description: Code review the current PR (or branch diff against main) for bugs, security, and AGENTS.md compliance. MUST use when asked to review code.
+description: Code review a pull request for bugs and CLAUDE.md compliance. MUST use when asked to review code.
 ---
 
-# Local Code Review
+# Local Code Review Skill
 
-Run the same review locally that the GitHub auto-review actions run on PRs (Claude / Codex / Pi). The review policy lives in `REVIEW.md`.
+Review a pull request for real bugs and CLAUDE.md compliance violations. This review targets HIGH SIGNAL issues only.
 
-**Why a subagent**: the review MUST run in a fresh context — not inline in the current session. If the user has been iterating on the diff, the main session has absorbed their reasoning and rationalizations, so it anchors and misses things CI catches. A subagent starts cold, like CI does.
+## Review Philosophy
 
-## Steps
+- **Only flag issues you are certain about.** If you are not sure an issue is real, do not flag it. False positives erode trust and waste reviewer time.
+- Think like a senior engineer doing a final review — flag things that would cause incidents, not things that are merely imperfect.
 
-1. **Determine the PR scope** (cheap, do this in the main session):
-   - If an argument is provided, treat it as a PR number or branch.
-   - Otherwise, detect from the current branch vs `main`.
-   - Confirm the PR/branch exists (`gh pr view <n>` or `git rev-parse <branch>`).
+## What to Flag
 
-2. **Delegate the review to a fresh-context subagent** with a self-contained prompt. The prompt MUST include:
-   - The PR number or branch name to review.
-   - The instruction to read `REVIEW.md` first for the policy, then `AGENTS.md` files in directories touched by the diff.
-   - The exact output format (see below).
-   - Whether `--comment` was requested (so the subagent emits inline-comment payloads if needed).
-   - Any "Additional reviewer instructions" the user provided.
+- Code that won't compile or parse (syntax errors, type errors, missing imports)
+- Code that will definitely produce wrong results regardless of inputs
+- Clear, unambiguous CLAUDE.md violations (quote the exact rule being violated)
+- Security issues in introduced code (injection, auth bypass, data exposure)
+- Incorrect logic that will fail in production
 
-   - **Claude Code**: use the `Agent` tool with `subagent_type: branch-diff-reviewer` (read-only tools, purpose-built for this). If unavailable, fall back to `general-purpose`.
-   - **Codex / Pi**: if the CLI exposes a fresh-session subagent mechanism, use it. Otherwise tell the user to run the skill in a fresh CLI session and stop — running inline in the current session defeats the purpose.
+## What NOT to Flag
 
-3. **Receive the findings** from the subagent and relay them to the user verbatim. Do not re-summarize, re-judge, or filter — the whole point of fresh context is to surface what the main session would dismiss.
+- Code style or quality concerns
+- Potential issues that depend on specific inputs or runtime state
+- Subjective suggestions or improvements
+- Pre-existing issues not introduced by this PR
+- Pedantic nitpicks a senior engineer wouldn't flag
+- Issues a linter or type checker will catch
+- General quality concerns unless explicitly prohibited in CLAUDE.md
+- Issues silenced via lint ignore comments
 
-4. **Post comments if `--comment` was requested**: use the `gh` commands below with the subagent's output as the body. The main session does the posting because the subagent is read-only.
+## Execution Steps
 
-## Subagent prompt template
+1. **Determine the PR scope**:
+   - If an argument is provided, use it as the PR number or branch
+   - Otherwise, detect from the current branch vs main
+   - Run `gh pr view` if a PR exists, or use `git diff main...HEAD`
 
-```
-Review <PR #N | branch X> against main per the policy in REVIEW.md.
+2. **Find relevant CLAUDE.md files**:
+   - Read the root `CLAUDE.md`
+   - Check for CLAUDE.md files in directories containing changed files
 
-Steps:
-1. Read REVIEW.md (repo root) for the full policy: severity triage, public-surface
-   checklist, AGENTS.md compliance, test coverage assessment.
-2. Read AGENTS.md (repo root) and any AGENTS.md in directories touched by the diff.
-3. Get the diff: `gh pr diff <N>` (if PR) or `git diff main...<branch>`.
-4. Get context: `gh pr view <N>` (if PR) or `git log main..<branch> --oneline`.
-5. Read changed files only when the diff alone is insufficient to validate a finding.
-6. Self-validate each finding: "is this definitely a real issue a senior engineer
-   would flag?" Discard if uncertain.
-7. Output findings in the exact format below. Do not modify any files.
+3. **Get the diff and metadata**:
+   - `gh pr diff` or `git diff main...HEAD` for the full diff
+   - `gh pr view` or `git log main..HEAD --oneline` for context
 
-<paste output format from below>
+4. **Read changed files** where the diff alone is insufficient to understand context
 
-<if --comment requested:>
-Additionally emit a JSON array of inline comments suitable for the GitHub reviews
-API, one per finding that maps to a specific line:
-[{"path": "...", "line": N, "side": "RIGHT", "body": "[P1] ..."}, ...]
-```
+5. **Review for**:
+   - CLAUDE.md compliance — check each rule against the changed code
+   - Bugs and logic errors — will this code work correctly?
+   - Security issues — injection, auth, data exposure in new code
 
-## Output format
+6. **Self-validate each finding**: Before reporting, ask yourself:
+   - "Is this definitely a real issue, not a false positive?"
+   - "Would a senior engineer flag this in review?"
+   - If the answer to either is no, discard the finding
+
+7. **Output findings** to the terminal (default) or post as PR comments (with `--comment` flag)
+
+## Output Format
 
 ```
 ## Code review
 
-<verdict line per REVIEW.md>
-
 Found N issues:
 
-1. [P0|P1|P2] <description>
+1. <description> (<reason: CLAUDE.md adherence | bug | security>)
    <file_path:line_number>
 
-2. [P0|P1|P2] <description>
+2. <description> (<reason>)
    <file_path:line_number>
 ```
 
-End with a `Test coverage` section per the shared policy.
-
 If no issues are found:
 
 ```
 ## Code review
 
-Good to merge.
-
-No issues found. Checked for bugs, security, and AGENTS.md compliance.
+No issues found. Checked for bugs and CLAUDE.md compliance.
 ```
 
-## Posting comments (`--comment`)
+## Posting Comments (--comment flag)
 
-For a top-level PR comment:
+If the user passes `--comment`, post findings as inline PR comments using:
 
 ```bash
-gh pr review --comment --body "<summary from subagent>"
+gh pr review --comment --body "<summary>"
 ```
 
-For inline comments on specific lines (using the JSON the subagent emitted):
+Or for inline comments on specific lines:
 
 ```bash
-gh api repos/{owner}/{repo}/pulls/{pr}/reviews \
-  -f body="<summary>" -f event="COMMENT" -f comments="<json from subagent>"
+gh api repos/{owner}/{repo}/pulls/{pr}/reviews -f body="<summary>" -f event="COMMENT" -f comments="[...]"
 ```

.agents/skills/native-trigger/SKILL.md

@@ -607,18 +607,7 @@ In `frontend/src/lib/components/triggers/TriggersEditor.svelte`:
 
 Add your service to the `nativeTriggerServices` map in `deleteDeployedTrigger()`. Native triggers use `NativeTriggerService.deleteNativeTrigger({ workspace, serviceName, externalId })` instead of the standard `path`-based delete.
 
-### Step 17: Update `getUsedTriggers` for Sidebar Visibility
-
-The sidebar (`frontend/src/lib/components/sidebar/SidebarContent.svelte`) shows native-trigger links only if `$usedTriggerKinds` includes the service — without this, your trigger page will never appear in the nav bar even when triggers exist.
-
-1. **Backend** — add `{service}_used: bool` to the `UsedTriggers` struct and SELECT in `backend/windmill-api-workspaces/src/workspaces.rs::get_used_triggers()`:
-   ```rust
-   EXISTS(SELECT 1 FROM native_trigger WHERE workspace_id = $1 AND service_name = '{service}'::native_trigger_service) AS "{service}_used!"
-   ```
-2. **OpenAPI** — add `{service}_used: boolean` to the response schema for `GET /w/{workspace}/workspaces/used_triggers` (under both `properties` and `required`).
-3. **Layout** — in `frontend/src/routes/(root)/(logged)/+layout.svelte::loadUsedTriggerKinds()`, destructure `{service}_used` and push `'{service}'` to `usedKinds`.
-
-### Step 18: Update OpenAPI Spec and Regenerate Types
+### Step 17: Update OpenAPI Spec and Regenerate Types
 
 Add to `JobTriggerKind` enum in `backend/windmill-api/openapi.yaml`, then:
 

.agents/skills/pr/SKILL.md

@@ -1,19 +1,17 @@
 ---
 name: pr
-user_invocable: true
-description: Open a draft pull request on GitHub and drive CI review rounds until it is ready. MUST use when you want to create/open a PR.
+description: Open a draft pull request on GitHub. MUST use when you want to create/open a PR.
 ---
 
 # Pull Request Skill
 
-Create a draft pull request with a clear title and explicit description of changes, then drive it through CI review rounds to ready.
+Create a draft pull request with a clear title and explicit description of changes.
 
 ## Instructions
 
 1. **Analyze branch changes**: Understand all commits since diverging from main
 2. **Push to remote**: Ensure all commits are pushed
 3. **Create draft PR**: Always open as draft for review before merging
-4. **Drive review rounds**: trigger CI reviews on the draft and only flip to ready once every verdict is a go (see "Review rounds" below)
 
 ## PR Title Format
 
@@ -52,63 +50,22 @@ The body MUST be explicit about what changed. Structure:
 ## Test plan
 - [ ] <How to verify change 1>
 - [ ] <How to verify change 2>
+
+---
+Generated with [Claude Code](https://claude.com/claude-code)
 ```
 
-The harness/tooling that invoked the skill may add its own attribution trailer; the skill itself does not prescribe one.
-
-## Screenshots (required for frontend changes)
-
-If `git diff main...HEAD --name-only` matches `^frontend/`, the PR body **must** include
-screenshots of the affected UI. Skip only when there is no visible UI effect (types,
-tests, build config) — and say so in the body.
-
-1. Verify the change in the browser (AGENTS.md → "Verifying Frontend Changes").
-2. Screenshot each affected page with `mcp__playwright__browser_take_screenshot` (save to a file).
-3. Host each image and get its Markdown embed by pushing to the public
-   `windmill-labs/agent-screenshots-internal` repo. **Pipe base64 through stdin** —
-   passing it as `-f content=…` fails with `argument list too long` on real images:
-
-   ```bash
-   REPO=windmill-labs/agent-screenshots-internal
-   IMG=screenshot.png                                          # repeat per page
-   DEST="shots/$(git branch --show-current)/$(date +%s)-$(basename "$IMG")"
-   base64 -w0 "$IMG" | jq -Rs --arg m "add $DEST" '{message:$m, content:.}' \
-     | gh api -X PUT "repos/$REPO/contents/$DEST" --input - >/dev/null
-   echo "![$(basename "$IMG" .png)](https://raw.githubusercontent.com/$REPO/main/$DEST)"
-   ```
-   Derive `$DEST` from the file name (as above) so distinct pages never collide — a
-   fixed name would make same-second uploads reuse one path, and the second `PUT`
-   then 422s (the Contents API needs the existing file's `sha` to overwrite).
-4. Put the printed `![…](…)` lines under a `## Screenshots` heading in the PR body.
-
-Requires `gh` (`repo` scope), `jq`, `base64` — all in the devShell. The host repo is
-public (so the raw URLs render for reviewers without a token) and its history is
-permanent — **never screenshot pages that show secrets or sensitive values** (workspace
-variables, resource values, instance settings, OAuth/SMTP config); deleting the file
-can't undo an accidental capture. (GitHub's drag-and-drop uploader needs a browser
-session and can't be driven from a token.)
-
-If `gh` can't push to the host repo (e.g. a CI token scoped only to `windmill`), do
-**not** fail the PR or skip silently — hand the upload to the user, who has push access,
-and continue once they confirm it's done.
-
 ## Execution Steps
 
 1. Run `git status` to check for uncommitted changes
 2. Run `git log main..HEAD --oneline` to see all commits in this branch
 3. Run `git diff main...HEAD` to see the full diff against main
-4. **Review the diff before creating the PR — run both reviews, do not skip:**
-   - **`local-review`** — Claude-native branch-diff-reviewer (`/local-review` in Claude Code, `$local-review` in Codex, `pi --skill local-review` / `/skill:local-review` in Pi).
-   - **`local-review-codex`** — cold Codex pass, the same review CI runs, for an independent perspective the Claude pass misses (`/local-review-codex` in Claude Code, or `bash .agents/skills/local-review-codex/run.sh`). If the `codex` CLI is missing or older than the version pinned in that skill, note it in your summary and continue — never block the PR on codex being unavailable.
-
-   Run both — they catch different things. If either surfaces issues, fix them and commit before proceeding.
-5. **Screenshots for frontend changes**: if `git diff main...HEAD --name-only` matches `^frontend/`, capture and embed screenshots of the affected UI per "Screenshots" above before writing the PR body (skip only if there is no visible UI effect).
-6. Check if remote branch exists and is up to date:
+4. Check if remote branch exists and is up to date:
    ```bash
    git rev-parse --abbrev-ref --symbolic-full-name @{u} 2>/dev/null || echo "no upstream"
    ```
-7. Push to remote if needed: `git push -u origin HEAD`
-8. Create draft PR using gh CLI:
+5. Push to remote if needed: `git push -u origin HEAD`
+6. Create draft PR using gh CLI:
    ```bash
    gh pr create --draft --title "<type>: <description>" --body "$(cat <<'EOF'
    ## Summary
@@ -121,46 +78,13 @@ and continue once they confirm it's done.
    ## Test plan
    - [ ] <test 1>
    - [ ] <test 2>
+
+   ---
+   Generated with [Claude Code](https://claude.com/claude-code)
    EOF
    )"
    ```
-9. Return the PR URL to the user
-10. Drive the PR through CI review rounds to ready (see "Review rounds" below)
-
-## Review rounds (draft → ready)
-
-A PR leaves draft **only after a clean CI review round**. Never run `gh pr ready` before that.
-
-1. **Trigger a round and wait for it**: launch the waiter as a background Bash task (a round takes 10–30 min; you are woken when it exits — do not stop the session or poll in the foreground while it runs):
-
-   ```bash
-   bash .agents/skills/pr/review-round.sh <PR_NUMBER>
-   ```
-
-   It comments `/review` on the PR — which runs the Codex, Claude and Pi CI reviewers even on a draft — waits for the spawned `PR Review Commands` workflow run(s) to complete, then prints one verdict line per reviewer and saves the full review comments to files.
-
-   `/review` (and `/codex`) are **idempotent per head SHA**: if a running or successful review already covers the current head, they skip that agent and post nothing new — the waiter reads the existing verdict for that head, so a skipped agent is *not* a missing one. A cancelled/failed head run is re-run in place; a fresh run is launched only when nothing covers the head. So an unchanged-head re-review is a near no-op, not a new round — push a commit to get genuinely fresh reviews.
-
-2. **Judge the round.** Codex is mandatory; Claude, Pi and cubic count whenever they posted. Every review starts with one of the three `REVIEW.md` verdicts:
-   - Codex verdict missing → the round is void: the waiter warns only when the head has no green Codex run (cancelled/failed/absent — not merely skipped-because-already-reviewed). Comment `/codex` on the PR, which re-runs the interrupted run in place (or launches one if none exists), wait the same way, and judge again.
-   - Any **"Should address issues before merging"** → fix the P0/P1 findings (and the nits while you're there), commit, push, and start a new round (step 1).
-   - Only **"Mergeable, but should ideally address nits"** and/or **"Good to merge"** → fix the nits too; a nit that is wrong or genuinely not worth fixing may instead be dismissed by replying to the review comment with your reasoning. Push nit-only fixes without starting another full round.
-
-3. **Flip to ready with the marker comment.** The review workflows skip the redundant `ready_for_review`-triggered round when the PR author has posted a marker naming the current head SHA **and** the PR's latest Codex review *posted before the marker* has a non-blocking verdict (reviewer evidence — a bare marker with no round behind it, or one whose last pre-marker Codex verdict is "Should address issues", skips nothing). Keep the prefix exact and use the full 40-char SHA of the head you are flipping:
-   - every verdict was "Good to merge" (head unchanged since the round):
-
-     `✅ Review round clean @ <head-sha>`
-
-   - nit-only round, nits fixed or dismissed afterwards (head may have moved past the reviewed SHA — say so):
-
-     `✅ Review round clean @ <head-sha> — nit-only verdicts at <round-sha>; nits addressed in <commit sha(s)> / dismissed in review replies`
-
-   ```bash
-   gh pr comment <PR_NUMBER> --body "✅ Review round clean @ $(git rev-parse HEAD)"
-   gh pr ready <PR_NUMBER>
-   ```
-
-   If any P0/P1 finding is unaddressed or the head moved for reasons other than nit fixes, do **not** post the marker or flip — run another round instead.
+7. Return the PR URL to the user
 
 ## EE Companion PR (when `*_ee.rs` files were modified)
 
@@ -176,6 +100,9 @@ Follow the full EE PR workflow in `docs/enterprise.md`. The key PR-specific deta
    ```bash
    gh pr create --draft --repo windmill-labs/windmill-ee-private --title "<type>: <description>" --body "$(cat <<'EOF'
    Companion PR for windmill-labs/windmill#<PR_NUMBER>
+
+   ---
+   Generated with [Claude Code](https://claude.com/claude-code)
    EOF
    )"
    ```

.agents/skills/pr/review-round.sh

@@ -1,172 +0,0 @@
-#!/usr/bin/env bash
-# Trigger a CI review round on a PR and wait for it to finish.
-#
-# Usage: bash .agents/skills/pr/review-round.sh [PR_NUMBER]
-#   PR_NUMBER defaults to the current branch's PR.
-#
-# Comments `/review` on the PR (works on drafts), waits for the spawned
-# "PR Review Commands" workflow run(s) to complete, then prints one verdict
-# line per reviewer and saves each full review comment to a file. A round
-# takes 10-30 minutes: run this in the background and act on its output when
-# it exits, per the pr skill ("Review rounds").
-set -euo pipefail
-
-REPO=${REPO:-$(gh repo view --json nameWithOwner --jq .nameWithOwner)}
-PR=${1:-$(gh pr view --json number --jq .number)}
-
-# Timestamp of the trigger comment, straight from GitHub, so local clock skew
-# can't make the run/comment filters below miss part of the round.
-TRIGGER_TIME=$(gh api "repos/$REPO/issues/$PR/comments" -f body='/review' --jq .created_at)
-echo "Review round triggered on $REPO#$PR at $TRIGGER_TIME"
-
-# Retry wrapper for one-off gh/API hiccups: a 45-minute wait must not die on
-# a single transient failure.
-retry() {
-  local attempt
-  for attempt in 1 2 3; do
-    if "$@"; then return 0; fi
-    sleep 10
-  done
-  return 1
-}
-
-# The /review comment spawns one "PR Review Commands" run holding the
-# claude/codex/pi jobs. Runs aren't linked to a PR, so wait on every run of
-# that workflow created after the trigger: a concurrent round on another PR
-# can only delay the answer, never truncate it. Every issue comment on any PR
-# spawns a fast-completing parse run of the same workflow, so the round's own
-# run may briefly lag the listing while unrelated runs already show completed:
-# require the all-completed state to hold past a floor and across two
-# consecutive polls before trusting it.
-DEADLINE=$(( $(date +%s) + 45 * 60 ))
-NO_RUN_DEADLINE=$(( $(date +%s) + 5 * 60 ))
-MIN_WAIT_UNTIL=$(( $(date +%s) + 3 * 60 ))
-STABLE=0
-FAILURES=0
-while :; do
-  if RUNS=$(gh run list --repo "$REPO" --workflow=pr-review-commands.yml \
-      --created ">=$TRIGGER_TIME" --limit 100 --json status); then
-    FAILURES=0
-  else
-    FAILURES=$(( FAILURES + 1 ))
-    if [ "$FAILURES" -ge 5 ]; then
-      echo "ERROR: listing workflow runs failed $FAILURES times in a row; aborting the wait." >&2
-      exit 1
-    fi
-    echo "WARNING: listing workflow runs failed (attempt $FAILURES/5); retrying in 60s." >&2
-    sleep 60
-    continue
-  fi
-  TOTAL=$(jq length <<<"$RUNS")
-  PENDING=$(jq '[.[] | select(.status != "completed")] | length' <<<"$RUNS")
-  NOW=$(date +%s)
-  if [ "$TOTAL" -gt 0 ] && [ "$PENDING" -eq 0 ] && [ "$NOW" -gt "$MIN_WAIT_UNTIL" ]; then
-    STABLE=$(( STABLE + 1 ))
-    if [ "$STABLE" -ge 2 ]; then
-      break
-    fi
-  else
-    STABLE=0
-  fi
-  if [ "$TOTAL" -eq 0 ] && [ "$NOW" -gt "$NO_RUN_DEADLINE" ]; then
-    echo "ERROR: no 'PR Review Commands' run appeared within 5 minutes of the /review comment; check that the comment author has write access and the workflow is enabled." >&2
-    exit 1
-  fi
-  if [ "$NOW" -gt "$DEADLINE" ]; then
-    echo "WARNING: review round still pending after 45 minutes; reporting whatever has been posted so far." >&2
-    break
-  fi
-  sleep 60
-done
-
-# Head SHA at trigger time. `/review` is idempotent per head: it skips an agent a
-# running/successful review already covers, re-runs a cancelled/failed one in place on a
-# separate head-tied run, and launches fresh only when nothing covers the head. Verdict
-# reading below therefore keys off the head, not just the trigger timestamp.
-HEAD_SHA=$(retry gh api "repos/$REPO/pulls/$PR" --jq .head.sha)
-echo "Reviewing head $HEAD_SHA"
-
-# Newest non-skipped run of <workflow> tied to the head ("status conclusion"), or empty
-# when none exists. A re-run-in-place or an already-covering review resolves on such a
-# head-tied run — separate from the pr-review-commands run waited on above (a fresh
-# launch instead runs inside it, and posts after the trigger). A `skipped` run is the
-# draft/fork gate and produced no review, so it is ignored.
-head_run_state() {
-  gh run list --repo "$REPO" --workflow "$1" --commit "$HEAD_SHA" --limit 20 \
-    --json databaseId,status,conclusion \
-    --jq '[.[] | select(.conclusion != "skipped")] | sort_by(.databaseId) | last | if . then "\(.status) \(.conclusion // "-")" else empty end' 2>/dev/null || true
-}
-
-# A re-run-in-place review lands on a head-tied run that finishes after the fast
-# pr-review-commands run, so let those settle before reading verdicts.
-for wf in codex-pr-review.yml pi-pr-review.yml pr-ready-review.yml; do
-  while :; do
-    case "$(head_run_state "$wf")" in
-      ""|"completed "*) break ;;
-      *) if [ "$(date +%s)" -gt "$DEADLINE" ]; then break; fi; sleep 30 ;;
-    esac
-  done
-done
-
-OUT_DIR=$(mktemp -d -t review-round-XXXXXX)
-COMMENTS_RAW=$(retry gh api "repos/$REPO/issues/$PR/comments?per_page=100" --paginate)
-# Two views: comments from THIS round (after the trigger) and the full history. A fresh
-# launch posts after the trigger; an idempotent skip leaves the covering verdict in the
-# earlier run's comment, so fall back to history when that agent's head run is green.
-jq -s --arg t "$TRIGGER_TIME" '[.[][] | select(.created_at > $t)]' \
-  <<<"$COMMENTS_RAW" > "$OUT_DIR/comments.json"
-jq -s '[.[][]]' <<<"$COMMENTS_RAW" > "$OUT_DIR/comments-all.json"
-# cubic posts through the PR reviews API, not issue comments.
-REVIEWS_RAW=$(retry gh api "repos/$REPO/pulls/$PR/reviews?per_page=100" --paginate)
-jq -s --arg t "$TRIGGER_TIME" '[.[][] | select((.submitted_at // "") > $t)]' \
-  <<<"$REVIEWS_RAW" > "$OUT_DIR/pr-reviews.json"
-
-VERDICT_RE='(Good to merge|Mergeable, but should ideally address nits|Should address issues before merging)'
-
-body_by_header() { # <file> <header-substring>
-  jq -r --arg h "$2" '[.[] | select(.body // "" | contains($h))] | last | .body // empty' "$1"
-}
-body_by_login() { # <file> <login>
-  jq -r --arg l "$2" '[.[] | select(.user.login == $l)] | last | .body // empty' "$1"
-}
-head_ok() { [ "$(head_run_state "$1")" = "completed success" ]; }
-# Latest verdict for a reviewer: prefer this round's comment; if none and the reviewer's
-# head run succeeded (an idempotent /review skipped re-reviewing an already-green head),
-# fall back to the covering comment from the full history.
-verdict_body() { # <header|login> <value> <workflow>
-  local body
-  body=$("body_by_$1" "$OUT_DIR/comments.json" "$2")
-  if [ -z "$body" ] && head_ok "$3"; then
-    body=$("body_by_$1" "$OUT_DIR/comments-all.json" "$2")
-  fi
-  printf '%s' "$body"
-}
-report() { # <reviewer-name> <comment-body>
-  local name=$1 body=$2 verdict
-  if [ -z "$body" ]; then
-    echo "$name: (no review posted for this head)"
-    return
-  fi
-  printf '%s\n' "$body" > "$OUT_DIR/$name.md"
-  verdict=$(printf '%s\n' "$body" | grep -m1 -oE "${VERDICT_RE}.*" | sed 's/\*\*//g' || true)
-  echo "$name: ${verdict:-(review posted but no verdict line; read $OUT_DIR/$name.md)}"
-}
-
-echo
-echo "=== Review round verdicts for $REPO#$PR (head $HEAD_SHA) ==="
-CODEX_BODY=$(verdict_body header '## Codex Review' codex-pr-review.yml)
-report codex "$CODEX_BODY"
-report claude "$(verdict_body login 'claude[bot]' pr-ready-review.yml)"
-report pi "$(verdict_body header '## Pi Review' pi-pr-review.yml)"
-CUBIC_BODY=$(jq -r '[.[] | select(.user.login | test("^cubic(-dev-ai)?(\\[bot\\])?$"; "i"))] | last | .body // empty' \
-  "$OUT_DIR/pr-reviews.json")
-if [ -z "$CUBIC_BODY" ]; then
-  CUBIC_BODY=$(jq -r '[.[] | select(.user.login | test("^cubic(-dev-ai)?(\\[bot\\])?$"; "i"))] | last | .body // empty' \
-    "$OUT_DIR/comments.json")
-fi
-report cubic "$CUBIC_BODY"
-echo
-echo "Full round output: $OUT_DIR (comments.json, pr-reviews.json, one .md per reviewer)"
-if [ -z "$CODEX_BODY" ]; then
-  echo "WARNING: no Codex verdict for $HEAD_SHA - its head run is not green (cancelled/failed/absent, not merely skipped-because-already-reviewed). Re-trigger with a '/codex' PR comment (re-runs the interrupted run in place, or launches one) and wait again." >&2
-fi

.agents/skills/refine/SKILL.md

@@ -1,6 +1,5 @@
 ---
 name: refine
-user_invocable: true
 description: End-of-session reflection. Reviews friction encountered during the session and proposes updates to docs/ to capture lessons learned.
 ---
 

.agents/skills/svelte-frontend/SKILL.md

@@ -78,7 +78,3 @@ Use the Svelte MCP tools when working on Svelte code:
 2. **get-documentation**: Fetch relevant sections based on use_cases
 3. **svelte-autofixer**: MUST use on all Svelte code before finalizing — keep calling until no issues
 4. **playground-link**: Only after user confirms and code was NOT written to project files
-
-## Verifying in the Browser
-
-After changing Svelte code, use the **Playwright MCP** (`mcp__playwright__*`) to drive the running frontend and confirm the change works. See AGENTS.md → "Verifying Frontend Changes" for the full flow. Use `playwright` (headless) on devboxes; `playwright-headed` when a display is available.

.agents/skills/update-sqlx/SKILL.md

@@ -1,155 +0,0 @@
----
-name: update-sqlx
-description: How to safely update SQLx offline query cache. MUST use when SQL queries change.
----
-
-# SQLx Offline Query Cache
-
-Windmill uses `SQLX_OFFLINE=true` in CI, which requires all `sqlx::query!` / `sqlx::query_as!` macros to have matching cached query data in `backend/.sqlx/`.
-
-## When to Run
-
-Run after **adding or editing** a SQL query in Rust source. Without it, CI fails with:
-```
-error: `SQLX_OFFLINE=true` but there is no cached data for this query
-```
-
-**Do NOT run it when a change only *removes* queries.** The cache is already complete for
-CI; all that is left are orphaned entries, which are cosmetic and never break a build.
-Running `prepare` to tidy them risks destroying the cache for no gain. Delete them
-offline instead: for each `.sqlx/query-*.json`, normalize its `query` field (strip `\`
-line-continuations, collapse whitespace) and check whether it still appears in any `.rs`
-file. That detector reports ~48 false positives in a CE checkout — EE queries live in
-`*_ee.rs` symlinks it cannot read — so **filter to the tables your change touched** and
-delete only those.
-
-## Before You Run Anything
-
-1. **Back the cache up.** `prepare` deletes `.sqlx/` *before* regenerating, so any compile
-   failure leaves it gutted (observed: 2350 → 142 entries).
-   ```bash
-   cp -r backend/.sqlx /tmp/sqlx_backup    # restore with: rm -rf backend/.sqlx && cp -r /tmp/sqlx_backup backend/.sqlx
-   ```
-2. **Point `DATABASE_URL` at THIS worktree's database.** `prepare` compiles every
-   `sqlx::query!` against the **live** database. Another worktree's DB lacks your
-   migrations, so every new-table query fails and takes the cache down with it. The
-   symptom is `relation "<your_new_table>" does not exist` — that is a wrong
-   `DATABASE_URL`, not a broken query. See AGENTS.md → "Per-worktree ports and database".
-
-## Queries Inside Tests Need `--all-targets`, Which Fails In A CE Checkout
-
-`prepare` only caches queries in code it compiles, and `--workspace` alone does **not**
-compile test targets. A `sqlx::query!` inside `tests/*.rs` therefore gets no entry, and CI
-fails on the test target with the usual "no cached data" error even though the lib built
-clean. `SQLX_OFFLINE=true cargo check --workspace --all-targets` is what reproduces it.
-
-Adding `--all-targets` caches them — and, in a CE checkout, **aborts partway through**:
-`backend/tests/otel.rs` imports `windmill_common::otel_ee`, which exists only behind the
-`private` feature, so the compile dies after `prepare` has already emptied `.sqlx/`.
-Observed: 2435 → 4 entries, `error: cargo check failed with status: exit status: 101`.
-
-Do not fight it — the abort is a pre-existing EE gap, not something your change caused.
-Take the entries you need and put the backup back:
-
-```bash
-cd backend
-cp -r .sqlx /tmp/sqlx_backup
-ls /tmp/sqlx_backup | sort > /tmp/before.txt
-
-DATABASE_URL=<this worktree's db> \
-  cargo sqlx prepare --workspace -- --workspace --features all_sqlx_features --all-targets
-# expected to fail; it still wrote the entries it got to before dying
-
-ls .sqlx | sort > /tmp/after.txt
-mkdir -p /tmp/newq
-comm -13 /tmp/before.txt /tmp/after.txt | while read f; do cp ".sqlx/$f" /tmp/newq/; done
-
-rm -rf .sqlx && cp -r /tmp/sqlx_backup .sqlx && cp /tmp/newq/*.json .sqlx/
-```
-
-**Read every file in `/tmp/newq` before copying it in** — print each one's `query` field and
-confirm it is one of yours. The set is small (one per new test query), and anything else in
-there means the run got further than you think.
-
-Then verify both targets, since the lib passing says nothing about the tests:
-
-```bash
-SQLX_OFFLINE=true cargo check --workspace --features all_sqlx_features   # lib
-SQLX_OFFLINE=true cargo check -p <your-crate> --all-targets              # tests
-```
-
-## The Problem
-
-`cargo sqlx prepare --workspace` **deletes all existing cache files** and regenerates only the ones found in the current compilation. If you don't compile with every feature flag (especially `private` for EE files), you will **silently delete EE query caches**, breaking CI for enterprise tests.
-
-The standard `./update_sqlx.sh` script tries to compile with all features, but it often fails locally because the EE symlinks can be out of sync with `main`.
-
-## Safe Procedure
-
-Always preserve the existing EE caches from `origin/main`. Use this workflow:
-
-```bash
-cd backend
-
-# 1. Restore the full cache from main (includes EE caches)
-git checkout origin/main -- .sqlx/
-
-# 2. Run prepare with OSS features (what compiles locally)
-#    This regenerates OSS caches to match your code changes.
-cargo sqlx prepare --workspace -- --workspace --features all_sqlx_features
-
-# 3. Restore any EE caches that were deleted in step 2.
-#    These are files present in origin/main but missing after prepare.
-git ls-tree origin/main backend/.sqlx/ \
-  | awk '{print $4}' | sed 's|backend/\.sqlx/||' | sort > /tmp/main_files.txt
-
-find backend/.sqlx -name "*.json" -printf '%P\n' | sort > /tmp/current_files.txt
-
-comm -23 /tmp/main_files.txt /tmp/current_files.txt > /tmp/missing_files.txt
-
-while read f; do
-  git show "origin/main:backend/.sqlx/$f" > "backend/.sqlx/$f"
-done < /tmp/missing_files.txt
-
-# 4. Verify nothing was lost from main
-find backend/.sqlx -name "*.json" -printf '%P\n' | sort > /tmp/current_files.txt
-comm -23 /tmp/main_files.txt /tmp/current_files.txt | wc -l
-# Should output: 0
-```
-
-## If EE Compiles Locally
-
-If your EE repo happens to be in sync, you can use the full script (faster):
-
-```bash
-cd backend
-./update_sqlx.sh
-```
-
-But if it fails with EE compilation errors, use the safe procedure above.
-
-## What NOT to Do
-
-- **Never** run `cargo sqlx prepare --workspace` with only OSS features and commit the result — it will delete EE caches.
-- **Never** set `SQLX_OFFLINE=true` for local `cargo sqlx prepare` — use a live database per CLAUDE.md. (CI runs with `SQLX_OFFLINE=true`, which is why the cache must be complete.)
-- **Never** run `prepare` without a `.sqlx` backup, or against a `DATABASE_URL` you have not confirmed belongs to this worktree.
-- **Never** run `prepare` at all for a removal-only change.
-- **Never** skip the verification step (step 4 above).
-- **Never** leave a `--all-targets` run's output in place after it aborts — it is a
-  near-empty cache. Restore the backup and graft on only the entries you verified.
-
-Step 4 compares against `origin/main` because step 1 restored from it, so the two agree.
-If you did **not** run step 1 — auditing a branch's cache on its own, say — compare
-against `git merge-base HEAD origin/main` instead: `origin/main` advances, so its newer
-entries would read as losses on your branch.
-
-## Verification
-
-After committing, the diff against `origin/main` should show:
-- A few **new** cache files (for your changed queries)
-- A few **deleted** cache files (for old queries that no longer exist)
-- **Zero** net deletions from the EE cache set
-
-```bash
-git diff origin/main --stat backend/.sqlx/
-```

.claude/hooks/allow-fileops-in-tmp.sh

@@ -1,179 +0,0 @@
-#!/usr/bin/env bash
-# PreToolUse allowance for scratch file ops: auto-allow a single, plain, single-line
-# `mkdir` / `cp` / `mv` / `touch` / `chmod` / `tar` / `unzip` whose every path operand
-# resolves under /tmp. Anything else makes no decision (exit 0) and falls back to the normal
-# permission flow — where `Bash(mv:*)` and `Bash(chmod:*)` in the `ask` list prompt. A
-# PreToolUse `allow` overrides those ask rules, which is why this is a hook and not an allow
-# rule: permission rules match a command prefix, so they can only constrain the FIRST operand.
-# `cp /tmp/x ~/.zshrc` matches a `cp /tmp/` prefix, and requiring every operand is the point.
-#
-# Requiring the sources under /tmp too (not just the destination) keeps this from becoming a
-# read-exfiltration path around the `Read(**/.env)` / `Read(**/secrets/**)` deny rules: a copy
-# out of the project into /tmp would land the content somewhere `Read(/tmp/**)` allows.
-#
-# Deny-by-default tokenizing, in the same spirit as guard-rm-outside-tmp.sh: every path token
-# must consist only of alphanumerics and `. _ / -`. That set contains none of the characters
-# bash uses for quoting, expansion, or command separation ($ ` ~ { } ( ) ' " \ ; & | < >), nor
-# any glob character, so all of those forms fail by construction. `realpath -m` then resolves
-# `..` and existing symlinks, so `/tmp/link` pointing at /etc/passwd is caught.
-#
-# `tar` and `unzip` get their own parser: their write destination arrives as a flag VALUE
-# (`-C`, `-d`) rather than a positional, and a bundle like `-xzf` consumes the token after it.
-# Flags are an allowlist, not a denylist, so `-P` / `--absolute-names` — which turn off tar's
-# refusal to extract `..` and absolute member paths — defer rather than needing enumeration.
-# Extraction additionally requires an explicit destination under /tmp, or a cwd already under
-# /tmp, since otherwise members land in the project checkout.
-#
-# Residual risk accepted: an archive whose members include a symlink pointing out of /tmp
-# followed by a write through it can still escape, because tar applies member symlinks as it
-# extracts. The archive itself must be under /tmp to get here, so this is a hazard only for
-# archives fetched from an untrusted source into the scratch dir.
-#
-# Assumes GNU `realpath` (-m) and `jq`, both present in this repo's Linux dev env.
-set -uo pipefail
-
-input=$(cat)
-command -v jq >/dev/null 2>&1 || exit 0
-cmd=$(printf '%s' "$input" | jq -r '.tool_input.command // empty' 2>/dev/null)
-[ -z "$cmd" ] && exit 0
-cwd=$(printf '%s' "$input" | jq -r '.cwd // empty' 2>/dev/null)
-
-# A newline separates commands, and the tokenizer below only reads the first line — defer.
-case "$cmd" in *$'\n'*) exit 0 ;; esac
-
-read -r -a toks <<< "$cmd"
-
-# 0 iff the token is charset-safe and resolves to a path strictly inside /tmp.
-under_tmp() {
-  local t="$1" canon
-  # Globs never auto-allow. Bash expands them only after this hook has decided, so realpath
-  # sees the unexpanded pattern: `/tmp/link*` canonicalizes to itself and passes, then
-  # expands onto a symlink whose target is outside /tmp. chmod and cp follow command-line
-  # symlinks, so that is a write to the target. guard-rm-outside-tmp.sh can allow globs
-  # because `rm` unlinks the symlink itself rather than following it.
-  case "$t" in *[*?[]*) return 1 ;; esac
-  [ -n "$(printf '%s' "$t" | tr -d 'A-Za-z0-9._/-')" ] && return 1
-  # Absolute only. Resolving a relative operand against the cwd makes any bare word look like
-  # a safe path whenever the cwd is under /tmp, while the tool itself reads it as an option:
-  # `tar P -xf ...` is --absolute-names, not ./P, and `cp /tmp/t -RL /tmp/o` is a
-  # dereferencing recursive copy, not a file named -RL.
-  case "$t" in /*) ;; *) return 1 ;; esac
-  canon=$(realpath -m -- "$t" 2>/dev/null)
-  [ -n "$canon" ] || return 1
-  # /tmp itself is never a target — only paths strictly inside it.
-  case "$canon" in /tmp/?*) return 0 ;; esac
-  return 1
-}
-
-allow() {
-  jq -nc --arg r "$1" '{hookSpecificOutput:{hookEventName:"PreToolUse",permissionDecision:"allow",permissionDecisionReason:$r}}'
-  exit 0
-}
-
-# Bare command word only; wrappers (`timeout cp`), env prefixes, and `/bin/cp` defer.
-# Options are an allowlist per command, so anything that changes how symlinks are followed
-# defers instead of needing enumeration. `cp -L` / `-H` matter most: they dereference while
-# recursing, which copies the CONTENT of a symlink target from outside /tmp into a scratch
-# dir that `Read(/tmp/**)` then exposes. Plain `-r` and `-a` (which implies `-d`) recreate
-# such a symlink as a symlink instead, so no outside content is materialized.
-case "${toks[0]:-}" in
-  mkdir) takes_mode=0; ok_opts='pv' ;;
-  cp)    takes_mode=0; ok_opts='rRvfnpa' ;;
-  mv)    takes_mode=0; ok_opts='vfn' ;;
-  touch) takes_mode=0; ok_opts='acmv' ;;
-  chmod) takes_mode=1; ok_opts='Rvfc' ;;   # chmod's first operand is a mode, not a path
-  tar)   ok_flags='xctzjJavfC'; val_flags='fC' ;;
-  unzip) ok_flags='oqnljvd';    val_flags='d'  ;;
-  *) exit 0 ;;
-esac
-
-# ---------------------------------------------------------------- tar / unzip
-if [ -n "${ok_flags:-}" ]; then
-  saw_archive=0 saw_dest=0 extracting=0 listing=0 end_opts=0
-  i=1
-  while [ "$i" -lt "${#toks[@]}" ]; do
-    t="${toks[$i]}"
-    i=$((i + 1))
-    if [ "$end_opts" = 0 ]; then
-      [ "$t" = "--" ] && { end_opts=1; continue; }
-      case "$t" in
-        -?*)
-          flags="${t#-}"
-          # Allowlist: a long option, -P/--absolute-names, --transform, -I and friends all
-          # leave a residue here and defer rather than being enumerated as denials.
-          [ -n "$(printf '%s' "$flags" | tr -d "$ok_flags")" ] && exit 0
-          case "$flags" in *x*) extracting=1 ;; esac
-          case "${toks[0]}$flags" in unzip*[lv]*) listing=1 ;; esac
-          # A flag consuming the next token must be alone in its bundle's final position
-          # (`-xzf a.tar`), else the token it eats is ambiguous.
-          case "${flags%?}" in *[$val_flags]*) exit 0 ;; esac
-          case "${flags: -1}" in
-            [$val_flags])
-              val="${toks[$i]:-}"
-              i=$((i + 1))
-              [ -n "$val" ] || exit 0
-              under_tmp "$val" || exit 0
-              case "${flags: -1}" in
-                f) saw_archive=1 ;;
-                C | d) saw_dest=1 ;;
-              esac
-              ;;
-          esac
-          continue
-          ;;
-      esac
-    fi
-    # Positional. For tar these are sources (create) or member names (extract); for unzip the
-    # first is the archive. Requiring every one under /tmp is conservative for member names,
-    # which are not filesystem paths — those defer rather than being wrongly allowed.
-    under_tmp "$t" || exit 0
-    [ "${toks[0]}" = "unzip" ] && saw_archive=1
-  done
-
-  [ "$saw_archive" = 1 ] || exit 0   # tar without -f reads a tape/stdin; unzip needs an archive
-  # Writes land relative to the working directory unless a destination was given. `unzip -l`
-  # and `-v` only list, so they need no destination.
-  if [ "$extracting" = 1 ] || { [ "${toks[0]}" = "unzip" ] && [ "$listing" = 0 ]; }; then
-    [ "$saw_dest" = 1 ] || under_tmp "${cwd:-$PWD}" || exit 0
-  fi
-  allow "archive paths and extraction target are under /tmp"
-fi
-
-# ------------------------------------------- mkdir / cp / mv / touch / chmod
-path_operand=0
-seen_mode=0
-end_opts=0
-i=1
-while [ "$i" -lt "${#toks[@]}" ]; do
-  t="${toks[$i]}"
-  i=$((i + 1))
-
-  if [ "$end_opts" = 0 ]; then
-    [ "$t" = "--" ] && { end_opts=1; continue; }
-    # Checked at any position, not just before the first operand: GNU utils permute, so
-    # `cp /tmp/tree -RL /tmp/out` still enables dereferencing recursion.
-    case "$t" in
-      -?*)
-        # Allowlist: long options and the dereferencing flags leave a residue and defer.
-        [ -n "$(printf '%s' "${t#-}" | tr -d "$ok_opts")" ] && exit 0
-        continue
-        ;;
-    esac
-  fi
-
-  # chmod: consume the mode operand without a path check. Octal, or symbolic clauses.
-  if [ "$takes_mode" = 1 ] && [ "$seen_mode" = 0 ]; then
-    case "$t" in
-      [0-7] | [0-7][0-7] | [0-7][0-7][0-7] | [0-7][0-7][0-7][0-7]) ;;
-      *) printf '%s' "$t" | grep -Eq '^[ugoa]*[+=-][rwxXst]*(,[ugoa]*[+=-][rwxXst]*)*$' || exit 0 ;;
-    esac
-    seen_mode=1
-    continue
-  fi
-
-  under_tmp "$t" || exit 0
-  path_operand=1
-done
-
-[ "$path_operand" = 1 ] || exit 0
-allow "every path operand is under /tmp"

.claude/hooks/format-frontend.sh

@@ -10,12 +10,8 @@ if [ -z "$FILE_PATH" ]; then
     exit 0
 fi
 
-# Only the frontend app itself, i.e. a "frontend" directory sitting at a repo root.
-# A bare */frontend/* substring also matches ai_evals/adapters/frontend and the
-# ai_evals app fixtures, which no prettier config governs — prettier then falls back
-# to its defaults and rewrites the whole file. Anchoring to $CLAUDE_PROJECT_DIR
-# instead would skip worktrees edited from a session rooted elsewhere.
-if [[ "$FILE_PATH" == *"/frontend/"* ]] && [[ -e "${FILE_PATH%%/frontend/*}/.git" ]]; then
+# Check if the file is in the frontend directory
+if [[ "$FILE_PATH" == *"/frontend/"* ]]; then
     # Check if it's a formattable file type
     if [[ "$FILE_PATH" =~ \.(ts|js|svelte|json|css|html|md)$ ]]; then
         cd "$CLAUDE_PROJECT_DIR/frontend" || exit 0

.claude/hooks/guard-main-branch.sh

@@ -16,23 +16,6 @@ command="$(echo "$input" | jq -r '.tool_input.command // empty')"
 if [[ "$command" =~ ^git\ (push|reset|revert|checkout|merge|rebase|commit|add) ]]; then
   branch="$(git rev-parse --abbrev-ref HEAD 2>/dev/null || true)"
   if [[ "$branch" == "main" ]]; then
-    echo "BLOCK: You are on the main branch. Create or switch to a feature branch first." >&2
-    exit 2
-  fi
-fi
-
-# Block force-push targeting main from any branch.
-if [[ "$command" =~ ^git[[:space:]]+push([[:space:]]|$) ]]; then
-  has_force=false
-  if [[ "$command" =~ (--force([[:space:]]|=|$)|--force-with-lease|[[:space:]]-f([[:space:]]|$)) ]]; then
-    has_force=true
-  fi
-  # `+ref` refspec syntax is also a force push.
-  if [[ "$command" =~ [[:space:]]\+[A-Za-z] ]]; then
-    has_force=true
-  fi
-  if $has_force && [[ "$command" =~ (^|[[:space:]:])\+?main([[:space:]]|$) ]]; then
-    echo "BLOCK: Force-push to main is not allowed via Claude. Run it yourself if you really mean to." >&2
-    exit 2
+    echo "BLOCK: You are on the main branch. Create or switch to a feature branch first."
   fi
 fi

.claude/hooks/guard-rm-outside-tmp.sh

@@ -1,106 +0,0 @@
-#!/usr/bin/env bash
-# PreToolUse guard for `rm`: auto-allow ONLY a single, plain, single-line `rm` whose every
-# operand is a whitelisted target — under /tmp, or inside a git working tree located in $HOME
-# (a version-controlled project dir). Anything else makes no decision (exit 0) and falls back
-# to the normal permission flow, where the `Bash(rm:*)` ask rule prompts (classifier as a
-# backstop).
-#
-# The git-tree allowance trades on "this is a project under version control" being lower-stakes
-# than a delete elsewhere — NOT on full recoverability: committed content is restorable via git,
-# but untracked / .gitignore'd / uncommitted content, and an independent nested repo's history
-# under a recursively-deleted parent, are NOT. Accepted as a deliberate convenience tradeoff.
-#
-# Deny-by-default: every token must consist only of a safe character set (alphanumerics,
-# `. _ / -` and glob chars `* ? [ ]`). That set contains none of the characters bash uses for
-# quoting, expansion, or command separation ($ ` ~ { } ( ) ' " \ ; & | < >), so those forms
-# fail by construction rather than needing to be enumerated. `realpath -m` then resolves `..`
-# and existing symlinks (so a symlink out of the allowed roots is caught), and a wildcard in a
-# non-final path segment is refused because it can expand through a symlink realpath can't see.
-#
-# The git-repo allowance covers targets inside a git working tree under $HOME, and the tree's
-# own root folder only when it is a linked worktree (`.git` is a pointer file, so history in
-# the main repo survives); a primary checkout's root (`.git` is a history dir) and any `.git`
-# path are never auto-allowed. Globs auto-allow only under /tmp — elsewhere their expansion
-# could reach `.git` or a dotfile the literal checks never see. Relative operands resolve
-# against the command's cwd (from the hook input). A PreToolUse `allow` overrides the ask rule.
-#
-# Assumes GNU `realpath` (-m) and `jq`, both present in this repo's Linux dev env.
-set -uo pipefail
-
-input=$(cat)
-command -v jq >/dev/null 2>&1 || exit 0
-cmd=$(printf '%s' "$input" | jq -r '.tool_input.command // empty' 2>/dev/null)
-[ -z "$cmd" ] && exit 0
-cwd=$(printf '%s' "$input" | jq -r '.cwd // empty' 2>/dev/null)
-
-# A newline separates commands, and the tokenizer below only reads the first line — defer.
-case "$cmd" in *$'\n'*) exit 0 ;; esac
-
-read -r -a toks <<< "$cmd"
-# Bare leading `rm` only; wrappers (`timeout rm`), env prefixes, and `/bin/rm` defer.
-[ "${toks[0]:-}" = "rm" ] || exit 0
-
-# 0 (allow) iff the canonical path is an auto-allowable rm target: under /tmp, or strictly
-# inside a git working tree located under $HOME. The walk stops at $HOME, so a dotfiles repo at
-# ~ can't make all of $HOME deletable, and top-level ~ files stay protected.
-allowed_target() {
-  local canon="$1" d root=""
-  case "$canon" in /tmp/?*) return 0 ;; esac
-  [ -n "${HOME:-}" ] || return 1
-  case "$canon" in "$HOME"/?*) ;; *) return 1 ;; esac
-  case "$canon" in *"/.git" | *"/.git/"*) return 1 ;; esac   # protect history, not recoverable
-  d="$canon"
-  while [ "$d" != "/" ] && [ "$d" != "$HOME" ]; do
-    [ -e "$d/.git" ] && { root="$d"; break; }
-    d=$(dirname "$d")
-  done
-  [ -n "$root" ] || return 1              # not inside a git working tree under $HOME
-  if [ "$canon" = "$root" ]; then
-    # Deleting the repo root folder itself: allow only for a linked worktree, whose `.git` is
-    # a file/pointer so the history lives in the main repo and survives. A primary checkout's
-    # `.git` is a directory holding the history, so deleting it is unrecoverable — defer.
-    [ -f "$root/.git" ] && return 0
-    return 1
-  fi
-  return 0
-}
-
-had_operand=0
-end_opts=0
-i=1
-while [ "$i" -lt "${#toks[@]}" ]; do
-  t="${toks[$i]}"
-  i=$((i + 1))
-  # Whitelist every token (flags included, so an operator hidden in a flag like `-rf;rm`
-  # can't slip past): any character outside the safe set makes it unsafe to reason about.
-  [ -n "$(printf '%s' "$t" | tr -d 'A-Za-z0-9._/*?[]-')" ] && exit 0
-  # A glob in an option-looking token (`-[-]`) can expand to `--` and turn a later `-name`
-  # into an operand — never a real option, so defer.
-  case "$t" in -*[*?[]*) exit 0 ;; esac
-  if [ "$end_opts" = 0 ]; then
-    [ "$t" = "--" ] && { end_opts=1; continue; }
-    # Skip real options only before the first operand. A bare `-` is a filename, and under
-    # POSIXLY_CORRECT GNU rm stops option parsing at the first operand, so a later `-name`
-    # is a filename too — validate it rather than skipping it.
-    if [ "$had_operand" = 0 ]; then
-      case "$t" in -?*) continue ;; esac
-    fi
-  fi
-  had_operand=1
-  # No wildcard in a non-final path segment (`a/*/b`): it can expand through a symlink
-  # realpath can't see. A slashless glob (`*.rs`) is a final-segment match — fine.
-  case "$t" in */*) case "${t%/*}" in *[*?[]*) exit 0 ;; esac ;; esac
-  case "$t" in
-    /*) canon=$(realpath -m -- "$t" 2>/dev/null) ;;
-    *)  canon=$(realpath -m -- "${cwd:-$PWD}/$t" 2>/dev/null) ;;
-  esac
-  [ -n "$canon" ] || exit 0
-  # A glob may auto-allow only under /tmp, where everything is deletable. Elsewhere its
-  # expansion could match `.git`, a dotfile like `.*`, or a nested checkout root that the
-  # literal-path checks never see — so require literal operands in git repos.
-  case "$t" in *[*?[]*) case "$canon" in /tmp/?*) ;; *) exit 0 ;; esac ;; esac
-  allowed_target "$canon" || exit 0
-done
-
-[ "$had_operand" = 1 ] || exit 0
-jq -nc '{hookSpecificOutput:{hookEventName:"PreToolUse",permissionDecision:"allow",permissionDecisionReason:"rm operands are under /tmp or inside a git checkout in $HOME"}}'

.claude/review-prompt.md

@@ -1,4 +1,25 @@
-# Claude output format
+# Code Review Instructions
 
-- Use inline comments at the relevant lines for specific issues.
-- Use a top-level comment for the summary, severity-tagged finding list, AGENTS.md compliance check, and the test-coverage assessment.
+Review this pull request and provide comprehensive feedback.
+
+## Focus Areas
+
+- **Code quality and best practices** — does the code follow established patterns?
+- **Potential bugs or issues** — will this code work correctly in all cases?
+- **Performance considerations** — are there unnecessary allocations, N+1 queries, or bottlenecks?
+- **Security implications** — injection, auth bypass, data exposure?
+
+## CLAUDE.md Compliance
+
+Read all relevant CLAUDE.md files (root and in directories containing changed files). Check each rule against the changed code. Quote the exact rule when flagging a violation.
+
+## Review Guidelines
+
+- Provide detailed feedback using inline comments for specific issues
+- Use top-level comments for general observations or praise
+- Only flag issues introduced by this PR, not pre-existing problems
+- Self-validate each finding: "Is this definitely a real issue?" If uncertain, discard it
+
+## Testing Instructions
+
+At the end of your review, add complete instructions to reproduce the added changes through the app interface. These instructions will be given to a tester so they can verify the changes. It should be a short descriptive text (not a step-by-step or a list) on how to navigate the app (what page, what action, what input, etc.) to see the changes.

.claude/settings.json

@@ -44,15 +44,7 @@
       "Bash(git merge:*)",
       "Bash(git rebase:*)",
       "Bash(git add:*)",
-      "Bash(git commit:*)",
-      "Read(/tmp/**)",
-      "Write(/tmp/**)",
-      "Edit(/tmp/**)",
-      "mcp__claude_ai_Gmail__search_threads",
-      "mcp__claude_ai_Gmail__get_thread",
-      "mcp__claude_ai_Gmail__get_message",
-      "mcp__claude_ai_Gmail__list_labels",
-      "mcp__claude_ai_Gmail__list_drafts"
+      "Bash(git commit:*)"
     ],
     "deny": [
       "Read(.env)",
@@ -63,10 +55,7 @@
       "Read(**/*.pem)",
       "Read(**/*.key)",
       "Read(**/credentials.json)",
-      "Read(**/.secret*)",
-      "Read(**/.secrets*)",
-      "Read(**/*.secret)",
-      "Read(**/*.secrets)",
+      "Read(**/*secret*)",
       "Edit(.env)",
       "Edit(.env.*)",
       "Edit(**/.env)",
@@ -80,21 +69,7 @@
       "Bash(chown:*)",
       "Bash(truncate:*)",
       "Bash(shred:*)",
-      "Bash(unlink:*)",
-      "mcp__claude_ai_Stripe",
-      "mcp__claude_ai_Gmail__create_draft",
-      "mcp__claude_ai_Gmail__update_draft",
-      "mcp__claude_ai_Gmail__create_label",
-      "mcp__claude_ai_Gmail__label_message",
-      "mcp__claude_ai_Gmail__label_thread",
-      "mcp__claude_ai_Gmail__unlabel_message",
-      "mcp__claude_ai_Gmail__unlabel_thread",
-      "mcp__claude_ai_Gmail__apply_sensitive_message_label",
-      "mcp__claude_ai_Gmail__apply_sensitive_thread_label",
-      "mcp__claude_ai_Google_Calendar",
-      "mcp__claude_ai_Google_Drive",
-      "mcp__claude_ai_Slack",
-      "mcp__claude_ai_Linear"
+      "Bash(unlink:*)"
     ]
   },
   "enableAllProjectMcpServers": true,
@@ -107,16 +82,6 @@
             "type": "command",
             "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/guard-main-branch.sh",
             "timeout": 5
-          },
-          {
-            "type": "command",
-            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/guard-rm-outside-tmp.sh",
-            "timeout": 5
-          },
-          {
-            "type": "command",
-            "command": "\"$CLAUDE_PROJECT_DIR\"/.claude/hooks/allow-fileops-in-tmp.sh",
-            "timeout": 5
           }
         ]
       }

.claude/skills/adding-a-trigger/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/adding-a-trigger/SKILL.md

.claude/skills/ai-chat/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/ai-chat/SKILL.md

.claude/skills/ai-evals/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/ai-evals/SKILL.md

.claude/skills/commit/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/commit/SKILL.md

.claude/skills/commit/SKILL.md

@@ -0,0 +1,60 @@
+---
+name: commit
+user_invocable: true
+description: Create a git commit with conventional commit format. MUST use anytime you want to commit changes.
+---
+
+# Git Commit Skill
+
+Create a focused, single-line commit following conventional commit conventions.
+
+## Instructions
+
+1. **Analyze changes**: Run `git status` and `git diff` to understand what was modified
+2. **Stage only modified files**: Add files individually by name. NEVER use `git add -A` or `git add .`
+3. **Write commit message**: Follow the conventional commit format as a single line
+
+## Conventional Commit Format
+
+```
+<type>: <description>
+```
+
+### Types
+- `feat`: New feature or capability
+- `fix`: Bug fix
+- `refactor`: Code change that neither fixes a bug nor adds a feature
+- `docs`: Documentation only changes
+- `style`: Formatting, missing semicolons, etc (no code change)
+- `test`: Adding or correcting tests
+- `chore`: Maintenance tasks, dependency updates, etc
+- `perf`: Performance improvement
+
+### Rules
+- Message MUST be a single line (no multi-line messages)
+- Description should be lowercase, imperative mood ("add" not "added")
+- No period at the end
+- Keep under 72 characters total
+
+### Examples
+```
+feat: add token usage tracking for AI providers
+fix: resolve null pointer in job executor
+refactor: extract common validation logic
+docs: update API endpoint documentation
+chore: upgrade sqlx to 0.7
+```
+
+## Execution Steps
+
+1. Run `git status` to see all changes
+2. Run `git diff` to understand the changes in detail
+3. Run `git log --oneline -5` to see recent commit style
+4. Stage ONLY the modified/relevant files: `git add <file1> <file2> ...`
+5. Create the commit with conventional format:
+   ```bash
+   git commit -m "<type>: <description>
+
+   Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>"
+   ```
+6. Run `git status` to verify the commit succeeded

.claude/skills/local-review-codex/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/local-review-codex/SKILL.md

.claude/skills/local-review/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/local-review/SKILL.md

.claude/skills/local-review/SKILL.md

@@ -0,0 +1,69 @@
+---
+name: local-review
+user_invocable: true
+description: Code review a pull request for bugs and CLAUDE.md compliance. MUST use when asked to review code.
+---
+
+# Local Code Review Skill
+
+Run the same review locally that the GitHub Claude Auto Review action runs on PRs. The shared review instructions live in `.claude/review-prompt.md` — read that file first and follow its instructions.
+
+## Execution Steps
+
+1. **Read `.claude/review-prompt.md`** for the review criteria and focus areas
+
+2. **Determine the PR scope**:
+   - If an argument is provided, use it as the PR number or branch
+   - Otherwise, detect from the current branch vs main
+   - Run `gh pr view` if a PR exists, or use `git diff main...HEAD`
+
+3. **Get the diff and metadata**:
+   - `gh pr diff` or `git diff main...HEAD` for the full diff
+   - `gh pr view` or `git log main..HEAD --oneline` for context
+
+4. **Read changed files** where the diff alone is insufficient to understand context
+
+5. **Apply the review instructions from `.claude/review-prompt.md`**
+
+6. **Self-validate each finding**: Before reporting, ask yourself:
+   - "Is this definitely a real issue, not a false positive?"
+   - "Would a senior engineer flag this in review?"
+   - If the answer to either is no, discard the finding
+
+7. **Output findings** to the terminal (default) or post as PR comments (with `--comment` flag)
+
+## Output Format
+
+```
+## Code review
+
+Found N issues:
+
+1. <description> (<reason: CLAUDE.md adherence | bug | security>)
+   <file_path:line_number>
+
+2. <description> (<reason>)
+   <file_path:line_number>
+```
+
+If no issues are found:
+
+```
+## Code review
+
+No issues found. Checked for bugs and CLAUDE.md compliance.
+```
+
+## Posting Comments (--comment flag)
+
+If the user passes `--comment`, post findings as inline PR comments using:
+
+```bash
+gh pr review --comment --body "<summary>"
+```
+
+Or for inline comments on specific lines:
+
+```bash
+gh api repos/{owner}/{repo}/pulls/{pr}/reviews -f body="<summary>" -f event="COMMENT" -f comments="[...]"
+```

.claude/skills/native-trigger/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/native-trigger/SKILL.md

.claude/skills/native-trigger/SKILL.md

@@ -0,0 +1,782 @@
+---
+name: native-trigger
+description: Guidance for adding native trigger services to Windmill. Use when implementing or modifying native trigger integrations across the backend and frontend.
+---
+
+# Skill: Adding Native Trigger Services
+
+This skill provides comprehensive guidance for adding new native trigger services to Windmill. Native triggers allow external services (like Nextcloud, Google Drive, etc.) to trigger Windmill scripts/flows via webhooks or push notifications.
+
+## Architecture Overview
+
+The native trigger system consists of:
+
+1. **Database Layer** - PostgreSQL tables and enum types
+2. **Backend Rust Implementation** - Core trait, handlers, and service modules in the `windmill-native-triggers` crate
+3. **Frontend Svelte Components** - Configuration forms and UI components
+
+### Key Files
+
+| Component | Path |
+|-----------|------|
+| Core module with `External` trait | `backend/windmill-native-triggers/src/lib.rs` |
+| Generic CRUD handlers | `backend/windmill-native-triggers/src/handler.rs` |
+| Background sync logic | `backend/windmill-native-triggers/src/sync.rs` |
+| OAuth/workspace integration | `backend/windmill-native-triggers/src/workspace_integrations.rs` |
+| Re-export shim (windmill-api) | `backend/windmill-api/src/native_triggers/mod.rs` |
+| TriggerKind enum | `backend/windmill-common/src/triggers.rs` |
+| JobTriggerKind enum | `backend/windmill-common/src/jobs.rs` |
+| Frontend service registry | `frontend/src/lib/components/triggers/native/utils.ts` |
+| Frontend trigger utilities | `frontend/src/lib/components/triggers/utils.ts` |
+| Trigger badges (icons + counts) | `frontend/src/lib/components/graph/renderers/triggers/TriggersBadge.svelte` |
+| Workspace integrations UI | `frontend/src/lib/components/workspaceSettings/WorkspaceIntegrations.svelte` |
+| OAuth config form component | `frontend/src/lib/components/workspaceSettings/OAuthClientConfig.svelte` |
+| OpenAPI spec | `backend/windmill-api/openapi.yaml` |
+| Reference: Nextcloud module | `backend/windmill-native-triggers/src/nextcloud/` |
+| Reference: Google module | `backend/windmill-native-triggers/src/google/` |
+
+### Crate Structure
+
+The native trigger code lives in the `windmill-native-triggers` crate (`backend/windmill-native-triggers/`). The `windmill-api` crate re-exports everything via a shim:
+
+```rust
+// backend/windmill-api/src/native_triggers/mod.rs
+pub use windmill_native_triggers::*;
+```
+
+All new service modules go in `backend/windmill-native-triggers/src/`.
+
+---
+
+## Core Concepts
+
+### The `External` Trait
+
+Every native trigger service implements the `External` trait defined in `lib.rs`:
+
+```rust
+#[async_trait]
+pub trait External: Send + Sync + 'static {
+    // Associated types:
+    type ServiceConfig: Debug + DeserializeOwned + Serialize + Send + Sync;
+    type TriggerData: Debug + Serialize + Send + Sync;
+    type OAuthData: DeserializeOwned + Serialize + Clone + Send + Sync;
+    type CreateResponse: DeserializeOwned + Send + Sync;
+
+    // Constants:
+    const SUPPORT_WEBHOOK: bool;
+    const SERVICE_NAME: ServiceName;
+    const DISPLAY_NAME: &'static str;
+    const TOKEN_ENDPOINT: &'static str;
+    const REFRESH_ENDPOINT: &'static str;
+    const AUTH_ENDPOINT: &'static str;
+
+    // Required methods:
+    async fn create(&self, w_id, oauth_data, webhook_token, data, db, tx) -> Result<Self::CreateResponse>;
+    async fn update(&self, w_id, oauth_data, external_id, webhook_token, data, db, tx) -> Result<serde_json::Value>;
+    async fn get(&self, w_id, oauth_data, external_id, db, tx) -> Result<Self::TriggerData>;
+    async fn delete(&self, w_id, oauth_data, external_id, db, tx) -> Result<()>;
+    async fn exists(&self, w_id, oauth_data, external_id, db, tx) -> Result<bool>;
+    async fn maintain_triggers(&self, db, workspace_id, triggers, oauth_data, synced, errors);
+    fn external_id_and_metadata_from_response(&self, resp) -> (String, Option<serde_json::Value>);
+
+    // Methods with defaults:
+    async fn prepare_webhook(&self, db, w_id, headers, body, script_path, is_flow) -> Result<PushArgsOwned>;
+    fn service_config_from_create_response(&self, data, resp) -> Option<serde_json::Value>;
+    fn additional_routes(&self) -> axum::Router;
+    async fn http_client_request<T, B>(&self, url, method, workspace_id, tx, db, headers, body) -> Result<T>;
+}
+```
+
+Key design points:
+- **`update()` returns `serde_json::Value`** - the resolved service_config to store. Each service is responsible for building the final config.
+- **`maintain_triggers()`** - periodic background maintenance. Each service implements its own strategy (Nextcloud: reconcile with external state; Google: renew expiring channels).
+- **No `list_all()` in the trait** - services that need it (Nextcloud) implement it privately; services that don't (Google) use different maintenance strategies.
+- **No `get_external_id_from_trigger_data()` or `extract_service_config_from_trigger_data()`** - removed in favor of the `maintain_triggers` pattern.
+
+### Create Lifecycle: Two Paths
+
+The `create_native_trigger` handler in `handler.rs` supports two creation flows, controlled by `service_config_from_create_response()`:
+
+**Path A: Short (Google pattern)** - `service_config_from_create_response()` returns `Some(config)`:
+1. `create()` registers on external service
+2. `external_id_and_metadata_from_response()` extracts the ID
+3. `service_config_from_create_response()` builds the config directly from input data + response metadata
+4. Stores trigger in DB -- done, no extra round-trip
+
+Use this when the external_id is known before the create call (e.g., Google generates the channel_id as a UUID upfront and includes it in the webhook URL).
+
+**Path B: Long (Nextcloud pattern)** - `service_config_from_create_response()` returns `None` (default):
+1. `create()` registers on external service (webhook URL has no external_id yet)
+2. `external_id_and_metadata_from_response()` extracts the ID
+3. `update()` is called to fix the webhook URL with the now-known external_id
+4. `update()` returns the resolved service_config
+5. Stores trigger in DB
+
+Use this when the external_id is assigned by the remote service and the webhook URL needs to be corrected after creation.
+
+### OAuth Token Storage (Three-Table Pattern)
+
+OAuth tokens are stored across three tables, NOT in `workspace_integrations.oauth_data` directly:
+
+| Table | What's Stored |
+|-------|---------------|
+| `workspace_integrations` | `oauth_data` JSON with `base_url`, `client_id`, `client_secret`, `instance_shared` flag; `resource_path` pointing to the variable |
+| `variable` | Encrypted `access_token` (at the path stored in `resource_path`), linked to `account` via `account` column |
+| `account` | `refresh_token`, keyed by `workspace_id` + `client` (service name) + `is_workspace_integration = true` |
+
+The `decrypt_oauth_data()` function in `lib.rs` assembles these into a unified struct:
+```rust
+pub struct OAuthConfig {
+    pub base_url: String,
+    pub access_token: String,      // decrypted from variable
+    pub refresh_token: Option<String>, // from account table
+    pub client_id: String,         // from oauth_data or instance settings
+    pub client_secret: String,     // from oauth_data or instance settings
+}
+```
+
+Instance-level sharing: when `oauth_data.instance_shared == true`, `client_id` and `client_secret` are read from global settings instead of workspace_integrations.
+
+### URL Resolution
+
+The `resolve_endpoint()` helper handles both absolute and relative OAuth URLs:
+
+```rust
+pub fn resolve_endpoint(base_url: &str, endpoint: &str) -> String {
+    if endpoint.starts_with("http://") || endpoint.starts_with("https://") {
+        endpoint.to_string()  // Google: absolute URLs
+    } else {
+        format!("{}{}", base_url, endpoint)  // Nextcloud: relative paths
+    }
+}
+```
+
+### ServiceName Methods
+
+`ServiceName` is the central registry enum. Each variant must implement these match arms:
+
+| Method | Purpose |
+|--------|---------|
+| `as_str()` | Lowercase identifier (e.g., `"google"`) |
+| `as_trigger_kind()` | Maps to `TriggerKind` enum |
+| `as_job_trigger_kind()` | Maps to `JobTriggerKind` enum |
+| `token_endpoint()` | OAuth token endpoint (relative or absolute) |
+| `auth_endpoint()` | OAuth authorization endpoint |
+| `oauth_scopes()` | Space-separated OAuth scopes |
+| `resource_type()` | Resource type for token storage (e.g., `"gworkspace"`) |
+| `extra_auth_params()` | Extra OAuth params (e.g., Google needs `access_type=offline`, `prompt=consent`) |
+| `integration_service()` | Maps to the workspace integration service (usually `*self`) |
+| `TryFrom<String>` | Parse from string |
+| `Display` | Delegates to `as_str()` |
+
+---
+
+## Step-by-Step Implementation Guide
+
+### Step 1: Database Migration
+
+Create a new migration file: `backend/migrations/YYYYMMDDHHMMSS_newservice_trigger.up.sql`
+
+```sql
+-- Add the service to the native_trigger_service enum
+ALTER TYPE native_trigger_service ADD VALUE IF NOT EXISTS 'newservice';
+
+-- Add to TRIGGER_KIND enum (used for trigger tracking)
+ALTER TYPE TRIGGER_KIND ADD VALUE IF NOT EXISTS 'newservice';
+
+-- Add to job_trigger_kind enum (used for job tracking)
+ALTER TYPE job_trigger_kind ADD VALUE IF NOT EXISTS 'newservice';
+```
+
+Also create the corresponding down migration.
+
+### Step 2: Update windmill-common Enums
+
+#### `backend/windmill-common/src/triggers.rs`
+
+Add variant to `TriggerKind` enum, and update `to_key()` and `fmt()` implementations.
+
+#### `backend/windmill-common/src/jobs.rs`
+
+Add variant to `JobTriggerKind` enum and update the `Display` implementation.
+
+### Step 3: Backend Service Module
+
+Create a new directory: `backend/windmill-native-triggers/src/newservice/`
+
+#### `mod.rs` - Type Definitions
+
+```rust
+use serde::{Deserialize, Serialize};
+
+pub mod external;
+// pub mod routes; // Only if you need additional service-specific routes
+
+/// OAuth data deserialized from the three-table pattern.
+/// The actual structure is built by decrypt_oauth_data() from variable + account + workspace_integrations.
+#[derive(Debug, Clone, Deserialize, Serialize)]
+pub struct NewServiceOAuthData {
+    pub base_url: String,              // from workspace_integrations.oauth_data
+    pub access_token: String,          // decrypted from variable table
+    pub refresh_token: Option<String>, // from account table
+    // Note: client_id and client_secret are in OAuthConfig, not here
+    // unless the service needs them at runtime for API calls
+}
+
+/// Configuration provided by user when creating/updating a trigger.
+/// Stored as JSON in native_trigger.service_config.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct NewServiceConfig {
+    // Service-specific configuration fields
+    pub folder_path: String,
+    pub file_filter: Option<String>,
+}
+
+/// Data retrieved from the external service about a trigger.
+/// Returned by the get() method and shown in the UI.
+#[derive(Debug, Clone, Serialize, Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct NewServiceTriggerData {
+    pub folder_path: String,
+    pub file_filter: Option<String>,
+    // Fields that shouldn't affect service_config comparison should use #[serde(skip_serializing)]
+}
+
+/// Response from external service when creating a trigger/webhook.
+#[derive(Debug, Deserialize)]
+pub struct CreateTriggerResponse {
+    pub id: String,
+}
+
+/// Handler struct (stateless, used for routing)
+#[derive(Copy, Clone)]
+pub struct NewService;
+```
+
+#### `external.rs` - External Trait Implementation
+
+```rust
+use async_trait::async_trait;
+use reqwest::Method;
+use sqlx::PgConnection;
+use std::collections::HashMap;
+use windmill_common::{
+    error::{Error, Result},
+    BASE_URL, DB,
+};
+
+use crate::{
+    generate_webhook_service_url, External, NativeTrigger, NativeTriggerData, ServiceName,
+    sync::{SyncError, TriggerSyncInfo},
+};
+use super::{NewService, NewServiceConfig, NewServiceOAuthData, NewServiceTriggerData, CreateTriggerResponse};
+
+#[async_trait]
+impl External for NewService {
+    type ServiceConfig = NewServiceConfig;
+    type TriggerData = NewServiceTriggerData;
+    type OAuthData = NewServiceOAuthData;
+    type CreateResponse = CreateTriggerResponse;
+
+    const SERVICE_NAME: ServiceName = ServiceName::NewService;
+    const DISPLAY_NAME: &'static str = "New Service";
+    const SUPPORT_WEBHOOK: bool = true;
+    const TOKEN_ENDPOINT: &'static str = "/oauth/token";
+    const REFRESH_ENDPOINT: &'static str = "/oauth/token";
+    const AUTH_ENDPOINT: &'static str = "/oauth/authorize";
+
+    async fn create(
+        &self,
+        w_id: &str,
+        oauth_data: &Self::OAuthData,
+        webhook_token: &str,
+        data: &NativeTriggerData<Self::ServiceConfig>,
+        db: &DB,
+        tx: &mut PgConnection,
+    ) -> Result<Self::CreateResponse> {
+        let base_url = &*BASE_URL.read().await;
+
+        // external_id is None during create (we get it from the response)
+        let webhook_url = generate_webhook_service_url(
+            base_url, w_id, &data.script_path, data.is_flow,
+            None, Self::SERVICE_NAME, webhook_token,
+        );
+
+        let url = format!("{}/api/webhooks/create", oauth_data.base_url);
+        let payload = serde_json::json!({
+            "callback_url": webhook_url,
+            "folder_path": data.service_config.folder_path,
+        });
+
+        let response: CreateTriggerResponse = self
+            .http_client_request(&url, Method::POST, w_id, tx, db, None, Some(&payload))
+            .await?;
+
+        Ok(response)
+    }
+
+    /// Update returns the resolved service_config as JSON.
+    /// For services using the update+get pattern, call self.get() and serialize.
+    async fn update(
+        &self,
+        w_id: &str,
+        oauth_data: &Self::OAuthData,
+        external_id: &str,
+        webhook_token: &str,
+        data: &NativeTriggerData<Self::ServiceConfig>,
+        db: &DB,
+        tx: &mut PgConnection,
+    ) -> Result<serde_json::Value> {
+        let base_url = &*BASE_URL.read().await;
+
+        let webhook_url = generate_webhook_service_url(
+            base_url, w_id, &data.script_path, data.is_flow,
+            Some(external_id), Self::SERVICE_NAME, webhook_token,
+        );
+
+        let url = format!("{}/api/webhooks/{}", oauth_data.base_url, external_id);
+        let payload = serde_json::json!({
+            "callback_url": webhook_url,
+            "folder_path": data.service_config.folder_path,
+        });
+
+        let _: serde_json::Value = self
+            .http_client_request(&url, Method::PUT, w_id, tx, db, None, Some(&payload))
+            .await?;
+
+        // Fetch back the updated state to get the resolved config
+        let trigger_data = self.get(w_id, oauth_data, external_id, db, tx).await?;
+        serde_json::to_value(&trigger_data)
+            .map_err(|e| Error::InternalErr(format!("Failed to serialize trigger data: {}", e)))
+    }
+
+    async fn get(
+        &self,
+        w_id: &str,
+        oauth_data: &Self::OAuthData,
+        external_id: &str,
+        db: &DB,
+        tx: &mut PgConnection,
+    ) -> Result<Self::TriggerData> {
+        let url = format!("{}/api/webhooks/{}", oauth_data.base_url, external_id);
+        self.http_client_request::<_, ()>(&url, Method::GET, w_id, tx, db, None, None).await
+    }
+
+    async fn delete(
+        &self,
+        w_id: &str,
+        oauth_data: &Self::OAuthData,
+        external_id: &str,
+        db: &DB,
+        tx: &mut PgConnection,
+    ) -> Result<()> {
+        let url = format!("{}/api/webhooks/{}", oauth_data.base_url, external_id);
+        let _: serde_json::Value = self
+            .http_client_request::<_, ()>(&url, Method::DELETE, w_id, tx, db, None, None)
+            .await
+            .or_else(|e| match &e {
+                Error::InternalErr(msg) if msg.contains("404") => Ok(serde_json::Value::Null),
+                _ => Err(e),
+            })?;
+        Ok(())
+    }
+
+    async fn exists(
+        &self,
+        w_id: &str,
+        oauth_data: &Self::OAuthData,
+        external_id: &str,
+        db: &DB,
+        tx: &mut PgConnection,
+    ) -> Result<bool> {
+        match self.get(w_id, oauth_data, external_id, db, tx).await {
+            Ok(_) => Ok(true),
+            Err(Error::NotFound(_)) => Ok(false),
+            Err(e) => Err(e),
+        }
+    }
+
+    /// Background maintenance. Choose the right pattern for your service:
+    /// - For services with queryable external state: use reconcile_with_external_state()
+    /// - For channel-based services with expiration: implement renewal logic
+    async fn maintain_triggers(
+        &self,
+        db: &DB,
+        workspace_id: &str,
+        triggers: &[NativeTrigger],
+        oauth_data: &Self::OAuthData,
+        synced: &mut Vec<TriggerSyncInfo>,
+        errors: &mut Vec<SyncError>,
+    ) {
+        // Option A: Reconcile with external state (Nextcloud pattern)
+        // Fetch all triggers from external service and compare with DB
+        let external_triggers = match self.list_all(workspace_id, oauth_data, db).await {
+            Ok(triggers) => triggers,
+            Err(e) => {
+                errors.push(SyncError {
+                    resource_path: format!("workspace:{}", workspace_id),
+                    error_message: format!("Failed to list triggers: {}", e),
+                    error_type: "api_error".to_string(),
+                });
+                return;
+            }
+        };
+
+        // Convert to (external_id, config_json) pairs
+        let external_pairs: Vec<(String, serde_json::Value)> = external_triggers
+            .into_iter()
+            .map(|t| (t.id.clone(), serde_json::to_value(&t).unwrap_or_default()))
+            .collect();
+
+        crate::sync::reconcile_with_external_state(
+            db, workspace_id, Self::SERVICE_NAME, triggers, &external_pairs, synced, errors,
+        ).await;
+    }
+
+    fn external_id_and_metadata_from_response(
+        &self,
+        resp: &Self::CreateResponse,
+    ) -> (String, Option<serde_json::Value>) {
+        (resp.id.clone(), None)
+    }
+
+    // service_config_from_create_response: NOT overridden (returns None).
+    // This means the handler uses the update+get pattern after create.
+    // Override and return Some(...) to skip the update+get cycle (Google pattern).
+}
+
+impl NewService {
+    /// Private helper to list all triggers from the external service.
+    async fn list_all(
+        &self,
+        w_id: &str,
+        oauth_data: &<Self as External>::OAuthData,
+        db: &DB,
+    ) -> Result<Vec<<Self as External>::TriggerData>> {
+        // Implementation depends on the external service's API
+        todo!()
+    }
+}
+```
+
+### Step 4: Update lib.rs Registry
+
+In `backend/windmill-native-triggers/src/lib.rs`:
+
+```rust
+// Service modules - add new services here:
+#[cfg(feature = "native_trigger")]
+pub mod newservice;  // <-- Add this
+
+// ServiceName enum - add variant:
+pub enum ServiceName {
+    Nextcloud,
+    Google,
+    NewService,  // <-- Add this
+}
+
+// Then add match arms in ALL ServiceName methods:
+// as_str(), as_trigger_kind(), as_job_trigger_kind(), token_endpoint(),
+// auth_endpoint(), oauth_scopes(), resource_type(), extra_auth_params(),
+// integration_service(), TryFrom<String>, Display
+```
+
+### Step 5: Update handler.rs Routes
+
+In `backend/windmill-native-triggers/src/handler.rs`:
+
+```rust
+pub fn generate_native_trigger_routers() -> Router {
+    // ...
+    #[cfg(feature = "native_trigger")]
+    {
+        use crate::newservice::NewService;
+        return router
+            .nest("/nextcloud", service_routes(NextCloud))
+            .nest("/google", service_routes(Google))
+            .nest("/newservice", service_routes(NewService));  // <-- Add this
+    }
+    // ...
+}
+```
+
+### Step 6: Update sync.rs
+
+In `backend/windmill-native-triggers/src/sync.rs`:
+
+```rust
+pub async fn sync_all_triggers(db: &DB) -> Result<BackgroundSyncResult> {
+    // ...
+    #[cfg(feature = "native_trigger")]
+    {
+        use crate::newservice::NewService;
+
+        // ... existing service syncs ...
+
+        // New service sync
+        let (service_name, result) = sync_service_triggers(db, NewService).await;
+        total_synced += result.synced_triggers.len();
+        total_errors += result.errors.len();
+        service_results.insert(service_name, result);
+    }
+    // ...
+}
+```
+
+### Step 7: Frontend Service Registry
+
+In `frontend/src/lib/components/triggers/native/utils.ts`:
+
+Add to `NATIVE_TRIGGER_SERVICES`, `getTriggerIconName()`, and `getServiceIcon()`.
+
+### Step 8: Frontend Trigger Form Component
+
+Create: `frontend/src/lib/components/triggers/native/services/newservice/NewServiceTriggerForm.svelte`
+
+### Step 9: Frontend Icon Component
+
+Create: `frontend/src/lib/components/icons/NewServiceIcon.svelte`
+
+### Step 10: Update NativeTriggerEditor
+
+Check `frontend/src/lib/components/triggers/native/NativeTriggerEditor.svelte` to ensure it dynamically loads form components based on service name.
+
+### Step 11: Workspace Integration UI
+
+Add your service to the `supportedServices` map in `frontend/src/lib/components/workspaceSettings/WorkspaceIntegrations.svelte`:
+
+```typescript
+const supportedServices: Record<string, ServiceConfig> = {
+    // ... existing services ...
+    newservice: {
+        name: 'newservice',
+        displayName: 'New Service',
+        description: 'Connect to New Service for triggers',
+        icon: NewServiceIcon,
+        docsUrl: 'https://www.windmill.dev/docs/integrations/newservice',
+        requiresBaseUrl: false,  // false for cloud services, true for self-hosted
+        setupInstructions: [
+            'Step 1: Create an OAuth app on the service',
+            'Step 2: Configure the redirect URI shown below',
+            'Step 3: Enter the client credentials below'
+        ]
+    }
+}
+```
+
+### Step 12: Update `frontend/src/lib/components/triggers/utils.ts`
+
+Update ALL of these maps/functions:
+1. `triggerIconMap` - import and add icon
+2. `triggerDisplayNamesMap` - add display name
+3. `triggerTypeOrder` in `sortTriggers()` - add type
+4. `getLightConfig()` - add case for your service
+5. `getTriggerLabel()` - add case for your service
+6. `jobTriggerKinds` - add to array
+7. `countPropertyMap` - add count property
+8. `triggerSaveFunctions` - add save function
+
+### Step 13: Update TriggersBadge Component
+
+In `frontend/src/lib/components/graph/renderers/triggers/TriggersBadge.svelte`:
+
+1. Import the icon
+2. Add to `baseConfig` with `countKey` (the dynamic `availableNativeServices` loop does NOT set `countKey`)
+3. Add to the `allTypes` array
+
+### Step 14: Update TriggersWrapper.svelte
+
+In `frontend/src/lib/components/triggers/TriggersWrapper.svelte`:
+
+Add a `{:else if selectedTrigger.type === 'yourservice'}` case that renders `<NativeTriggersPanel service="yourservice" ...>` with the same props pattern as the existing native trigger cases (e.g., `nextcloud`).
+
+### Step 15: Update AddTriggersButton.svelte
+
+In `frontend/src/lib/components/triggers/AddTriggersButton.svelte`:
+
+1. Add `yourserviceAvailable` state variable
+2. Add `setYourserviceState()` async function using `isServiceAvailable('yourservice', $workspaceStore!)`
+3. Call it at module level
+4. Add a dropdown entry to `addTriggerItems` with `hidden: !yourserviceAvailable`
+
+### Step 16: Update TriggersEditor.svelte Delete Handling
+
+In `frontend/src/lib/components/triggers/TriggersEditor.svelte`:
+
+Add your service to the `nativeTriggerServices` map in `deleteDeployedTrigger()`. Native triggers use `NativeTriggerService.deleteNativeTrigger({ workspace, serviceName, externalId })` instead of the standard `path`-based delete.
+
+### Step 17: Update OpenAPI Spec and Regenerate Types
+
+Add to `JobTriggerKind` enum in `backend/windmill-api/openapi.yaml`, then:
+
+```bash
+cd frontend && npm run generate-backend-client
+```
+
+---
+
+## Special Patterns
+
+### Unified Service with `trigger_type` (Google Pattern)
+
+When a single service handles multiple trigger types (e.g., Google Drive + Calendar share OAuth and API patterns), use a single `ServiceName` variant with a discriminator field:
+
+```rust
+pub enum GoogleTriggerType { Drive, Calendar }
+
+pub struct GoogleServiceConfig {
+    pub trigger_type: GoogleTriggerType,
+    // Drive-specific fields (only used when trigger_type = Drive)
+    pub resource_id: Option<String>,
+    pub resource_name: Option<String>,
+    // Calendar-specific fields (only used when trigger_type = Calendar)
+    pub calendar_id: Option<String>,
+    pub calendar_name: Option<String>,
+    // Metadata set after creation
+    pub google_resource_id: Option<String>,
+    pub expiration: Option<String>,
+}
+```
+
+Branch in trait methods based on `trigger_type`. Frontend uses a `ToggleButtonGroup` to switch between types. This keeps the codebase simpler (one service, one OAuth flow, one set of routes).
+
+See `backend/windmill-native-triggers/src/google/` for the reference implementation.
+
+### Skipping update+get After Create (Google Pattern)
+
+Override `service_config_from_create_response()` to return `Some(config)` when the external_id is known before the create call:
+
+```rust
+fn service_config_from_create_response(
+    &self,
+    data: &NativeTriggerData<Self::ServiceConfig>,
+    resp: &Self::CreateResponse,
+) -> Option<serde_json::Value> {
+    // Clone input config, add metadata from response
+    let mut config = data.service_config.clone();
+    config.google_resource_id = Some(resp.resource_id.clone());
+    config.expiration = Some(resp.expiration.clone());
+    Some(serde_json::to_value(&config).unwrap())
+}
+```
+
+### Services with Absolute OAuth Endpoints (Google)
+
+Unlike self-hosted services where OAuth endpoints are relative paths appended to `base_url`, services like Google have absolute URLs:
+
+```rust
+// Nextcloud: relative paths
+ServiceName::Nextcloud => "/apps/oauth2/api/v1/token",
+// Google: absolute URLs
+ServiceName::Google => "https://oauth2.googleapis.com/token",
+```
+
+The `resolve_endpoint()` function handles both. For services with absolute endpoints:
+- `base_url` can be empty
+- `requiresBaseUrl: false` in the frontend workspace integration config
+- Add `extra_auth_params()` if needed (Google requires `access_type=offline` and `prompt=consent`)
+
+### Channel-Based Push Notifications with Renewal (Google Pattern)
+
+For services using expiring watch channels instead of persistent webhooks:
+
+1. Store expiration in `service_config` (as part of `ServiceConfig`)
+2. In `maintain_triggers()`, implement renewal logic instead of using `reconcile_with_external_state()`:
+   ```rust
+   async fn maintain_triggers(&self, db, workspace_id, triggers, oauth_data, synced, errors) {
+       for trigger in triggers {
+           if should_renew_channel(trigger) {
+               self.renew_channel(db, trigger, oauth_data).await;
+           }
+       }
+   }
+   ```
+3. Renewal: best-effort stop old channel, create new one with same external_id, update service_config with new expiration
+4. Google example: Drive channels expire in 24h (renew when <1h left), Calendar channels expire in 7 days (renew when <1 day left)
+
+### reconcile_with_external_state (Nextcloud Pattern)
+
+The reusable function in `sync.rs` compares external triggers with DB state:
+- Triggers missing externally: sets error "Trigger no longer exists on external service"
+- Triggers present externally: clears errors, updates service_config if it differs
+
+Usage in `maintain_triggers()`:
+```rust
+let external_pairs: Vec<(String, serde_json::Value)> = /* fetch from external */;
+crate::sync::reconcile_with_external_state(
+    db, workspace_id, Self::SERVICE_NAME, triggers, &external_pairs, synced, errors,
+).await;
+```
+
+### Webhook Payload Processing
+
+Override `prepare_webhook()` to parse service-specific payloads into script/flow args:
+
+```rust
+async fn prepare_webhook(&self, db, w_id, headers, body, script_path, is_flow) -> Result<PushArgsOwned> {
+    let mut args = HashMap::new();
+    args.insert("event_type".to_string(), Box::new(headers.get("x-event-type").cloned()) as _);
+    args.insert("payload".to_string(), Box::new(serde_json::from_str::<serde_json::Value>(&body)?) as _);
+    Ok(PushArgsOwned { extra: None, args })
+}
+```
+
+Then register in `prepare_native_trigger_args()` in `lib.rs`:
+```rust
+pub async fn prepare_native_trigger_args(service_name, db, w_id, headers, body) -> Result<Option<PushArgsOwned>> {
+    match service_name {
+        ServiceName::Google => { /* ... */ Ok(Some(args)) }
+        ServiceName::NewService => { /* ... */ Ok(Some(args)) }
+        ServiceName::Nextcloud => Ok(None), // Uses default body parsing
+    }
+}
+```
+
+### Instance-Level OAuth Credentials
+
+When `workspace_integrations.oauth_data.instance_shared == true`, `decrypt_oauth_data()` reads `client_id` and `client_secret` from instance-level global settings instead of workspace-level. This allows admins to share OAuth app credentials across workspaces.
+
+The frontend handles this via the `generate_instance_connect_url` endpoint in `workspace_integrations.rs`.
+
+---
+
+## Testing Checklist
+
+- [ ] Database migration runs successfully
+- [ ] `cargo check -p windmill-native-triggers --features native_trigger` passes
+- [ ] `npx svelte-check --threshold error` passes (in frontend/)
+- [ ] Service appears in workspace integrations list
+- [ ] OAuth flow completes successfully
+- [ ] Can create a new trigger
+- [ ] Can view trigger details
+- [ ] Can update trigger configuration
+- [ ] Can delete trigger
+- [ ] Webhook receives and processes payloads
+- [ ] Background sync works correctly (reconciliation or channel renewal)
+- [ ] Error handling works (expired tokens, service unavailable)
+
+---
+
+## Reference Implementations
+
+### Nextcloud (Self-Hosted, Update+Get Pattern)
+
+| File | Purpose |
+|------|---------|
+| `nextcloud/mod.rs` | Types: NextCloudOAuthData, NextcloudServiceConfig, NextCloudTriggerData |
+| `nextcloud/external.rs` | External trait: uses update+get pattern, reconcile_with_external_state for sync |
+| `nextcloud/routes.rs` | Additional route: `GET /events` |
+
+Key patterns: relative OAuth endpoints, base_url required, list_all + reconcile for sync, update returns JSON from get().
+
+### Google (Cloud, Unified Service, Short Create)
+
+| File | Purpose |
+|------|---------|
+| `google/mod.rs` | Types: GoogleServiceConfig with trigger_type discriminator, GoogleTriggerType enum |
+| `google/external.rs` | External trait: overrides service_config_from_create_response, channel renewal for sync |
+| `google/routes.rs` | Additional routes: `GET /calendars`, `GET /drive/files`, `GET /drive/shared_drives` |
+
+Key patterns: absolute OAuth endpoints, empty base_url, trigger_type for Drive/Calendar, expiring watch channels with renewal, service_config_from_create_response skips update+get, get() reconstructs data from stored service_config (no external "get channel" API).

.claude/skills/pr/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/pr/SKILL.md

.claude/skills/pr/SKILL.md

@@ -0,0 +1,111 @@
+---
+name: pr
+user_invocable: true
+description: Open a draft pull request on GitHub. MUST use when you want to create/open a PR.
+---
+
+# Pull Request Skill
+
+Create a draft pull request with a clear title and explicit description of changes.
+
+## Instructions
+
+1. **Analyze branch changes**: Understand all commits since diverging from main
+2. **Push to remote**: Ensure all commits are pushed
+3. **Create draft PR**: Always open as draft for review before merging
+
+## PR Title Format
+
+Follow conventional commit format for the PR title:
+```
+<type>: <description>
+```
+
+### Types
+- `feat`: New feature or capability
+- `fix`: Bug fix
+- `refactor`: Code restructuring
+- `docs`: Documentation changes
+- `chore`: Maintenance tasks
+- `perf`: Performance improvements
+
+### Title Rules
+- Keep under 70 characters
+- Use lowercase, imperative mood
+- No period at the end
+- If `*_ee.rs` files were modified, prefix with `[ee]`: `[ee] <type>: <description>`
+
+## PR Body Format
+
+The body MUST be explicit about what changed. Structure:
+
+```markdown
+## Summary
+<Clear description of what this PR does and why>
+
+## Changes
+- <Specific change 1>
+- <Specific change 2>
+- <Specific change 3>
+
+## Test plan
+- [ ] <How to verify change 1>
+- [ ] <How to verify change 2>
+
+---
+Generated with [Claude Code](https://claude.com/claude-code)
+```
+
+## Execution Steps
+
+1. Run `git status` to check for uncommitted changes
+2. Run `git log main..HEAD --oneline` to see all commits in this branch
+3. Run `git diff main...HEAD` to see the full diff against main
+4. **Run `/local-review`** before creating the PR. If issues are found, fix them and commit before proceeding. Do not skip this step.
+5. Check if remote branch exists and is up to date:
+   ```bash
+   git rev-parse --abbrev-ref --symbolic-full-name @{u} 2>/dev/null || echo "no upstream"
+   ```
+6. Push to remote if needed: `git push -u origin HEAD`
+7. Create draft PR using gh CLI:
+   ```bash
+   gh pr create --draft --title "<type>: <description>" --body "$(cat <<'EOF'
+   ## Summary
+   <description>
+
+   ## Changes
+   - <change 1>
+   - <change 2>
+
+   ## Test plan
+   - [ ] <test 1>
+   - [ ] <test 2>
+
+   ---
+   Generated with [Claude Code](https://claude.com/claude-code)
+   EOF
+   )"
+   ```
+8. Return the PR URL to the user
+
+## EE Companion PR (when `*_ee.rs` files were modified)
+
+The `*_ee.rs` files in the windmill repo are **symlinks** to `windmill-ee-private` — changes won't appear in `git diff` of the windmill repo. Instead, check the EE repo for uncommitted or unpushed changes.
+
+Follow the full EE PR workflow in `docs/enterprise.md`. The key PR-specific details:
+
+1. Find the EE repo/worktree: see "Finding the EE Repo" in `docs/enterprise.md`
+2. Check for changes: `git -C <ee-path> status --short`
+   - If there are no changes in the EE repo, skip this entire section
+3. Follow steps 1–5 from the "EE PR Workflow" in `docs/enterprise.md`
+4. Create the companion PR (title does NOT get the `[ee]` prefix):
+   ```bash
+   gh pr create --draft --repo windmill-labs/windmill-ee-private --title "<type>: <description>" --body "$(cat <<'EOF'
+   Companion PR for windmill-labs/windmill#<PR_NUMBER>
+
+   ---
+   Generated with [Claude Code](https://claude.com/claude-code)
+   EOF
+   )"
+   ```
+5. Commit `ee-repo-ref.txt` and push the updated windmill branch

.claude/skills/refine/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/refine/SKILL.md

.claude/skills/refine/SKILL.md

@@ -0,0 +1,39 @@
+---
+name: refine
+user_invocable: true
+description: End-of-session reflection. Reviews friction encountered during the session and proposes updates to docs/ to capture lessons learned.
+---
+
+# Refine Skill
+
+Reflect on the current session and update documentation with lessons learned.
+
+## Instructions
+
+1. **Identify friction**: Review what happened in this session:
+   - Run `git diff main...HEAD --stat` to see what files were touched
+   - Think about: what was slow, what failed, what required multiple attempts, what information was missing or hard to find
+
+2. **Read current docs**: Read the docs that were relevant to this session:
+   - `docs/validation.md`
+   - `docs/enterprise.md`
+   - `docs/autonomous-mode.md`
+   - Any skills that were invoked
+
+3. **Propose updates**: For each piece of friction, decide if it warrants a doc update:
+   - **Missing knowledge**: Information you had to discover that should be documented
+   - **Wrong guidance**: Instructions that led you astray
+   - **Missing validation rule**: A check that should be in the validation matrix
+   - **New pattern**: A codebase pattern worth capturing for next time
+
+4. **Apply updates**: Edit the relevant `docs/` files. Keep changes minimal and specific — add only what would have saved time this session.
+
+5. **Report**: Summarize what was added/changed and why.
+
+## Rules
+
+- Only add knowledge confirmed by this session — no speculative additions
+- Keep docs concise — add a line or two, not a paragraph
+- If a whole new doc is needed, create it in `docs/` and add a pointer in `CLAUDE.md`
+- Don't update skills unless a coding pattern was genuinely wrong
+- Don't add things Claude already knows — only Windmill-specific knowledge

.claude/skills/rust-backend/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/rust-backend/SKILL.md

.claude/skills/rust-backend/SKILL.md

@@ -0,0 +1,107 @@
+---
+name: rust-backend
+description: Rust coding guidelines for the Windmill backend. MUST use when writing or modifying Rust code in the backend directory.
+---
+
+# Windmill Rust Patterns
+
+Apply these Windmill-specific patterns when writing Rust code in `backend/`.
+
+## Error Handling
+
+Use `Error` from `windmill_common::error`. Return `Result<T, Error>` or `JsonResult<T>`:
+
+```rust
+use windmill_common::error::{Error, Result};
+
+pub async fn get_job(db: &DB, id: Uuid) -> Result<Job> {
+    sqlx::query_as!(Job, "SELECT id, workspace_id FROM v2_job WHERE id = $1", id)
+        .fetch_optional(db)
+        .await?
+        .ok_or_else(|| Error::NotFound("job not found".to_string()))?;
+}
+```
+
+Never panic in library code. Reserve `.unwrap()` for compile-time guarantees.
+
+## SQLx Patterns
+
+**Never use `SELECT *`** — always list columns explicitly. Critical for backwards compatibility when workers lag behind API version:
+
+```rust
+// Correct
+sqlx::query_as!(Job, "SELECT id, workspace_id, path FROM v2_job WHERE id = $1", id)
+
+// Wrong — breaks when columns are added
+sqlx::query_as!(Job, "SELECT * FROM v2_job WHERE id = $1", id)
+```
+
+Use batch operations to avoid N+1:
+
+```rust
+// Preferred — single query with IN clause
+sqlx::query!("SELECT ... WHERE id = ANY($1)", &ids[..]).fetch_all(db).await?
+```
+
+Use transactions for multi-step operations. Parameterize all queries.
+
+## JSON Handling
+
+Prefer `Box<serde_json::value::RawValue>` over `serde_json::Value` when storing/passing JSON without inspection:
+
+```rust
+pub struct Job {
+    pub args: Option<Box<serde_json::value::RawValue>>,
+}
+```
+
+Only use `serde_json::Value` when you need to inspect or modify the JSON.
+
+## Serde Optimizations
+
+```rust
+#[derive(Serialize, Deserialize)]
+pub struct Job {
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub parent_job: Option<Uuid>,
+    #[serde(skip_serializing_if = "Vec::is_empty")]
+    pub tags: Vec<String>,
+    #[serde(default)]
+    pub priority: i32,
+}
+```
+
+## Async & Concurrency
+
+Never block the async runtime. Use `spawn_blocking` for CPU-intensive work:
+
+```rust
+let result = tokio::task::spawn_blocking(move || expensive_computation(&data)).await?;
+```
+
+**Mutex selection**: Prefer `std::sync::Mutex` (or `parking_lot::Mutex`) for data protection. Only use `tokio::sync::Mutex` when holding locks across `.await` points.
+
+Use `tokio::sync::mpsc` (bounded) for channels. Avoid `std::thread::sleep` in async contexts.
+
+## Module Structure & Visibility
+
+- Use `pub(crate)` instead of `pub` when possible
+- Place new code in the appropriate crate based on functionality
+- API endpoints go in `windmill-api/src/` organized by domain
+- Shared functionality goes in `windmill-common/src/`
+
+## Code Navigation
+
+Always use rust-analyzer LSP for go-to-definition, find-references, and type info. Do not guess at module paths.
+
+## Axum Handlers
+
+Destructure extractors directly in function signatures:
+
+```rust
+async fn process_job(
+    Extension(db): Extension<DB>,
+    Path((workspace, job_id)): Path<(String, Uuid)>,
+    Query(pagination): Query<Pagination>,
+) -> Result<Json<Job>> { ... }
+```

.claude/skills/svelte-frontend/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/svelte-frontend/SKILL.md

.claude/skills/svelte-frontend/SKILL.md

@@ -0,0 +1,80 @@
+---
+name: svelte-frontend
+description: Svelte coding guidelines for the Windmill frontend. MUST use when writing or modifying code in the frontend directory.
+---
+
+# Windmill Svelte Patterns
+
+Apply these Windmill-specific patterns when writing Svelte code in `frontend/`. For general Svelte 5 syntax (runes, snippets, event handling), use the Svelte MCP server.
+
+## Windmill UI Components (MUST use)
+
+Always use Windmill's design-system components. Never use raw HTML elements.
+
+### Buttons — `<Button>`
+
+```svelte
+<script>
+  import { Button } from '$lib/components/common'
+  import { ChevronLeft } from 'lucide-svelte'
+</script>
+
+<Button variant="default" onclick={handleClick}>Label</Button>
+<Button startIcon={{ icon: ChevronLeft }} iconOnly onclick={prev} />
+```
+
+Props: `variant?: 'accent' | 'accent-secondary' | 'default' | 'subtle'`, `unifiedSize?: 'sm' | 'md' | 'lg'`, `startIcon?: { icon: SvelteComponent }`, `iconOnly?: boolean`, `disabled?: boolean`
+
+### Text inputs — `<TextInput>`
+
+```svelte
+<script>
+  import { TextInput } from '$lib/components/common'
+</script>
+
+<TextInput bind:value={val} placeholder="Enter value" />
+```
+
+Props: `value?: string | number` (bindable), `placeholder?: string`, `disabled?: boolean`, `error?: string | boolean`, `size?: 'sm' | 'md' | 'lg'`
+
+### Selects — `<Select>`
+
+```svelte
+<script>
+  import Select from '$lib/components/select/Select.svelte'
+</script>
+
+<Select items={[{ label: 'Jan', value: 1 }]} bind:value={selected} />
+```
+
+Props: `items?: Array<{ label?: string; value: any }>`, `value` (bindable), `placeholder?: string`, `clearable?: boolean`, `size?: 'sm' | 'md' | 'lg'`
+
+### Icons — `lucide-svelte`
+
+Never write inline SVGs. Import from `lucide-svelte`:
+
+```svelte
+<script>
+  import { ChevronLeft, X } from 'lucide-svelte'
+</script>
+<ChevronLeft size={16} />
+```
+
+## Form Components
+
+Form components (TextInput, Toggle, Select, etc.) should use the unified size system when placed together.
+
+## Styling
+
+- Use Tailwind CSS for all styling — no custom CSS
+- Use Windmill's theming classes for colors/surfaces (see `frontend/brand-guidelines.md`)
+- Read component props JSDoc before using them
+
+## Svelte MCP Server
+
+Use the Svelte MCP tools when working on Svelte code:
+
+1. **list-sections**: Call first to discover available docs
+2. **get-documentation**: Fetch relevant sections based on use_cases
+3. **svelte-autofixer**: MUST use on all Svelte code before finalizing — keep calling until no issues
+4. **playground-link**: Only after user confirms and code was NOT written to project files

.claude/skills/update-sqlx/SKILL.md

@@ -1 +0,0 @@
-../../../.agents/skills/update-sqlx/SKILL.md

.codex/config.toml

@@ -1,4 +0,0 @@
-#:schema https://developers.openai.com/codex/config-schema.json
-
-[mcp_servers.svelte]
-url = "https://mcp.svelte.dev/mcp"

.gitattributes

@@ -1,3 +0,0 @@
-# Files a generator owns. Collapsed in review diffs and left out of language
-# stats: reviewing them means reviewing the generator instead.
-*.gen.ts linguist-generated=true

.github/DockerfileBackendTests

@@ -28,7 +28,7 @@ ENV PATH="${PATH}:/usr/local/go/bin"
 ENV GO_PATH=/usr/local/go/bin/go
 
 # UV
-RUN curl --proto '=https' --tlsv1.2 -LsSf https://github.com/astral-sh/uv/releases/download/0.11.24/uv-installer.sh | sh && mv /usr/local/cargo/bin/uv /usr/local/bin/uv
+RUN curl --proto '=https' --tlsv1.2 -LsSf https://github.com/astral-sh/uv/releases/download/0.9.24/uv-installer.sh | sh && mv /usr/local/cargo/bin/uv /usr/local/bin/uv
 
 ENV TZ=Etc/UTC
 

.github/change-versions-mac.sh

@@ -7,7 +7,7 @@ VERSION=$1
 echo "Updating versions to: $VERSION"
 
 sed -i '' -e "/^version =/s/= .*/= \"$VERSION\"/" ${root_dirpath}/backend/Cargo.toml
-sed -i '' -e "/^export const VERSION =/s/= .*/= \"v$VERSION\";/"  ${root_dirpath}/cli/src/core/constants.ts
+sed -i '' -e "/^export const VERSION =/s/= .*/= \"v$VERSION\";/"  ${root_dirpath}/cli/src/main.ts
 sed -i '' -e "/^export const VERSION =/s/= .*/= \"v$VERSION\";/" ${root_dirpath}/benchmarks/lib.ts
 sed -i '' -e "/version: /s/: .*/: $VERSION/" ${root_dirpath}/backend/windmill-api/openapi.yaml
 sed -i '' -e "/version: /s/: .*/: $VERSION/" ${root_dirpath}/openflow.openapi.yaml
@@ -20,10 +20,4 @@ sed -i '' -e "/^wmill =/s/= .*/= \">=$VERSION\"/" ${root_dirpath}/lsp/Pipfile
 
 sed -i '' -E "s/name = \"windmill\"\nversion = \"[^\"]*\"\\n(.*)/name = \"windmill\"\nversion = \"$VERSION\"\\n\\1/" ${root_dirpath}/backend/Cargo.lock
 
-# windmill-parser-wasm is its own workspace (excluded from the backend workspace
-# because of nightly-only cargo-features), so its version lives in
-# [workspace.package] and its Cargo.lock is not regenerated by the backend step.
-sed -i '' -e "/^version =/s/= .*/= \"$VERSION\"/" ${root_dirpath}/backend/parsers/windmill-parser-wasm/Cargo.toml
-sed -i '' -E "s/(name = \"windmill[^\"]*\"\nversion = )\"[^\"]*\"/\\1\"$VERSION\"/g" ${root_dirpath}/backend/parsers/windmill-parser-wasm/Cargo.lock
-
 cd ${root_dirpath}/frontend && npm i --package-lock-only

.github/change-versions.sh

@@ -7,14 +7,13 @@ VERSION=$1
 echo "Updating versions to: $VERSION"
 
 sed -i -e "/^version =/s/= .*/= \"$VERSION\"/" ${root_dirpath}/backend/Cargo.toml
-sed -i -e "/^export const VERSION =/s/= .*/= \"$VERSION\";/" ${root_dirpath}/cli/src/core/constants.ts
+sed -i -e "/^export const VERSION =/s/= .*/= \"$VERSION\";/" ${root_dirpath}/cli/src/main.ts
 sed -i -e "/^export const VERSION =/s/= .*/= \"v$VERSION\";/" ${root_dirpath}/benchmarks/lib.ts
 sed -i -e "/version: /s/: .*/: $VERSION/" ${root_dirpath}/backend/windmill-api/openapi.yaml
 sed -i -e "/version: /s/: .*/: $VERSION/" ${root_dirpath}/openflow.openapi.yaml
 sed -i -e "/\"version\": /s/: .*,/: \"$VERSION\",/" ${root_dirpath}/typescript-client/package.json
 sed -i -e "/\"version\": /s/: .*,/: \"$VERSION\",/" ${root_dirpath}/typescript-client/jsr.json
 sed -i -e "/\"version\": /s/: .*,/: \"$VERSION\",/" ${root_dirpath}/frontend/package.json
-sed -i -e "/\"version\": /s/: .*,/: \"$VERSION\",/" ${root_dirpath}/windmill-yaml-validator/package.json
 sed -i -e "/^version =/s/= .*/= \"$VERSION\"/" ${root_dirpath}/python-client/wmill/pyproject.toml
 sed -i -e "/^windmill-api =/s/= .*/= \"\\^$VERSION\"/" ${root_dirpath}/python-client/wmill/pyproject.toml
 sed -i -e "/^[[:space:]]*ModuleVersion[[:space:]]*=/s/= .*/= '$VERSION'/" ${root_dirpath}/powershell-client/WindmillClient/WindmillClient.psd1
@@ -22,14 +21,4 @@ sed -i -e "/^wmill =/s/= .*/= \">=$VERSION\"/" ${root_dirpath}/lsp/Pipfile
 
 sed -i -zE "s/name = \"windmill\"\nversion = \"[^\"]*\"\\n(.*)/name = \"windmill\"\nversion = \"$VERSION\"\\n\\1/" ${root_dirpath}/backend/Cargo.lock
 
-# windmill-parser-wasm is its own workspace (excluded from the backend workspace
-# because of nightly-only cargo-features), so its version lives in
-# [workspace.package] and its Cargo.lock is not regenerated by the backend step.
-sed -i -e "/^version =/s/= .*/= \"$VERSION\"/" ${root_dirpath}/backend/parsers/windmill-parser-wasm/Cargo.toml
-sed -i -zE "s/(name = \"windmill[^\"]*\"\nversion = )\"[^\"]*\"/\\1\"$VERSION\"/g" ${root_dirpath}/backend/parsers/windmill-parser-wasm/Cargo.lock
-
 cd ${root_dirpath}/frontend && npm i --package-lock-only --ignore-scripts
-
-# The CLI installs this package on every `bun install`, which would otherwise rewrite the
-# lockfile's version and leave a dirty tree.
-cd ${root_dirpath}/windmill-yaml-validator && npm i --package-lock-only --ignore-scripts

.github/codex/pr-review.prompt.md

@@ -1,5 +1,23 @@
-# Codex output format
+You are reviewing a GitHub pull request for this repository.
 
-- Read the review context file whose absolute path is given at the end of these instructions; it holds the PR metadata and the diff commands.
-- Return a markdown PR comment starting with `## Codex Review`.
-- Tag each finding with a severity (P0 / P1 / P2), file path, and line number when known confidently.
+Review policy:
+- Read `CLAUDE.md` before reviewing code.
+- Only report issues you are confident are real and introduced by this pull request.
+- Focus on bugs, security problems, and clear `CLAUDE.md` violations.
+- Do not report style nits, speculative concerns, pre-existing issues, or problems that a normal linter/typechecker would obviously catch.
+- Keep the review high signal. If there is no clear issue, return no findings.
+
+Repository context:
+- Read `./.github/codex/pr-review-context.md` for the PR metadata and the exact diff commands to use.
+- Review only the changes introduced by this PR.
+- Read additional files only when the diff is not enough to validate a finding.
+- Do not modify any files.
+
+Output requirements:
+- Return a GitHub PR comment in markdown, not JSON.
+- Start with `## Codex Review`.
+- Give a short overall summary first.
+- If you found high-signal issues, list them in a short numbered list with file paths and line numbers when you know them confidently.
+- If you found no high-signal issues, say that explicitly.
+- End with a `### Reproduction instructions` section containing a short descriptive paragraph for a tester explaining how to navigate the app to observe the change. Do not make it a numbered list. If the diff is not enough to infer this safely, say that plainly.
+- Prefer at most 10 findings.

.github/pi/pr-review.prompt.md

@@ -1,6 +0,0 @@
-# Pi output format
-
-- Read the review context file whose absolute path is given at the end of these instructions; it holds the PR metadata and the diff (or the git commands to produce it).
-- Return a markdown PR comment starting with `## Pi Review`.
-- Tag each finding with a severity (P0 / P1 / P2), file path, and line number when known confidently.
-- Output ONLY the final review markdown — no preamble, no thinking, no tool transcripts.

.github/scripts/check-docs-links.mjs

@@ -1,160 +0,0 @@
-// Extracts every windmill.dev/docs link referenced in the frontend source and
-// verifies none of them 404. Run: `node .github/scripts/check-docs-links.mjs`.
-// Used by the check-docs-links GitHub workflow (release / manual trigger only).
-
-import { readdir, readFile } from 'node:fs/promises'
-import { join, extname } from 'node:path'
-
-const ROOT = 'frontend/src'
-const EXTS = new Set(['.ts', '.js', '.svelte', '.mjs', '.cjs'])
-const DOCS_RE = /https?:\/\/(?:www\.)?windmill\.dev\/docs\/[^\s"'`)>\]}]*/g
-// `const someBaseUrl = 'https://www.windmill.dev/docs/...'` used later as `${someBaseUrl}/foo`
-const BASE_RE = /(?:const|let|var)\s+([A-Za-z_$][\w$]*)\s*=\s*['"`](https?:\/\/(?:www\.)?windmill\.dev\/docs\/[^'"`]+)['"`]/g
-
-const CONCURRENCY = 24
-const TIMEOUT_MS = 20000
-const RETRIES = 2
-
-// Links whose target page is written but not yet deployed on windmill.dev: the app
-// link is already the final slug, so a 404 is expected until the docs side ships.
-// The value is why the entry exists, for whoever has to judge whether it still should.
-const PENDING_DEPLOY = new Map()
-
-async function walk(dir) {
-	const out = []
-	for (const entry of await readdir(dir, { withFileTypes: true })) {
-		const p = join(dir, entry.name)
-		if (entry.isDirectory()) {
-			if (entry.name === 'node_modules' || entry.name === '.svelte-kit') continue
-			out.push(...(await walk(p)))
-		} else if (EXTS.has(extname(entry.name))) {
-			out.push(p)
-		}
-	}
-	return out
-}
-
-// url (no fragment) -> Set of source files it appears in
-const urls = new Map()
-const unresolved = []
-
-function record(url, file) {
-	const clean = url
-		.replace(/\\.*$/, '') // cut at an escape sequence embedded in a string literal (e.g. \n)
-		.replace(/#.*$/, '') // drop anchor fragment — irrelevant to a 404 check
-		.replace(/[.,;:'")\]]+$/, '')
-	if (!clean) return
-	// A `{`/`${` means the URL is built from an unresolved template/interpolation var.
-	if (clean.includes('{')) {
-		unresolved.push(`${clean}  (${file})`)
-		return
-	}
-	if (!urls.has(clean)) urls.set(clean, new Set())
-	urls.get(clean).add(file)
-}
-
-for (const file of await walk(ROOT)) {
-	let content = await readFile(file, 'utf8')
-	// Inline file-local base-url constants so `${base}/page` template literals resolve.
-	const bases = []
-	for (const m of content.matchAll(BASE_RE)) bases.push({ name: m[1], value: m[2], decl: m[0] })
-	for (const { name, value } of bases) {
-		content = content.replaceAll('${' + name + '}', value)
-	}
-	// Blank each base declaration so a prefix-only base (no index page of its own,
-	// e.g. .../app_configuration_settings) isn't checked as a standalone link.
-	// A genuinely bare `${base}` usage was already inlined above, so it's still covered.
-	for (const { decl } of bases) content = content.replace(decl, '')
-	for (const m of content.matchAll(DOCS_RE)) record(m[0], file)
-}
-
-const allUrls = [...urls.keys()].sort()
-console.log(`Found ${allUrls.length} distinct docs links across ${ROOT}`)
-if (unresolved.length) {
-	console.log(`\n⚠️  ${unresolved.length} link(s) built from an unrecognized base URL — skipped (register the base const so they can be checked):`)
-	for (const u of [...new Set(unresolved)].sort()) console.log(`   ${u}`)
-}
-
-async function check(url) {
-	for (let attempt = 0; attempt <= RETRIES; attempt++) {
-		const ctrl = new AbortController()
-		const timer = setTimeout(() => ctrl.abort(), TIMEOUT_MS)
-		try {
-			let res = await fetch(url, {
-				method: 'HEAD',
-				redirect: 'follow',
-				signal: ctrl.signal,
-				headers: { 'user-agent': 'windmill-docs-link-check' }
-			})
-			// Some hosts reject HEAD — fall back to GET.
-			if (res.status === 405 || res.status === 501) {
-				res = await fetch(url, {
-					method: 'GET',
-					redirect: 'follow',
-					signal: ctrl.signal,
-					headers: { 'user-agent': 'windmill-docs-link-check' }
-				})
-			}
-			clearTimeout(timer)
-			return { url, status: res.status, ok: res.status < 400 }
-		} catch (err) {
-			clearTimeout(timer)
-			if (attempt === RETRIES) return { url, status: 0, ok: false, error: String(err?.message || err) }
-			await new Promise((r) => setTimeout(r, 500 * (attempt + 1)))
-		}
-	}
-}
-
-// Simple concurrency pool.
-const results = []
-let idx = 0
-async function worker() {
-	while (idx < allUrls.length) {
-		const url = allUrls[idx++]
-		results.push(await check(url))
-	}
-}
-await Promise.all(Array.from({ length: CONCURRENCY }, worker))
-
-// An entry claims one thing — the page is not published yet — and 404 is the only
-// answer that means it. A timeout, 403 or 5xx on the same URL is a real fault, and
-// suppressing it would also read as "still waiting" and defer the staleness check.
-const isPendingDeploy = (r) => PENDING_DEPLOY.has(r.url) && r.status === 404
-
-const pending = results.filter((r) => PENDING_DEPLOY.has(r.url))
-const waiting = results.filter(isPendingDeploy)
-if (waiting.length) {
-	console.log(`\n⏳ ${waiting.length} link(s) waiting on a docs deploy:`)
-	for (const p of waiting.sort((a, b) => a.url.localeCompare(b.url))) {
-		console.log(`   ${p.url}\n     ${PENDING_DEPLOY.get(p.url)} — not live yet (${p.status})`)
-	}
-}
-
-// An entry that outlived its reason exempts a URL from the check forever, so a stale
-// one has to fail the job: a line in a green log is not read at release time.
-const stale = [
-	...pending.filter((p) => p.ok).map((p) => [p.url, 'the page is live']),
-	...[...PENDING_DEPLOY.keys()].filter((u) => !urls.has(u)).map((u) => [u, 'nothing references it'])
-]
-
-const failures = results.filter((r) => !r.ok && !isPendingDeploy(r))
-if (failures.length === 0 && stale.length === 0) {
-	console.log(`\n✅ No broken docs links (${allUrls.length} checked).`)
-	process.exit(0)
-}
-
-if (failures.length) {
-	console.log(`\n❌ ${failures.length} broken docs link(s):`)
-	for (const f of failures.sort((a, b) => a.url.localeCompare(b.url))) {
-		console.log(`\n  ${f.url}`)
-		console.log(`    status: ${f.error ? `error (${f.error})` : f.status}`)
-		for (const file of urls.get(f.url)) console.log(`    ↳ ${file}`)
-	}
-}
-if (stale.length) {
-	console.log(`\n❌ ${stale.length} PENDING_DEPLOY entr(ies) to delete from this script:`)
-	for (const [url, why] of stale.sort((a, b) => a[0].localeCompare(b[0]))) {
-		console.log(`\n  ${url}\n    ${why}`)
-	}
-}
-process.exit(1)

.github/workflows/ai-agent-tests.yml

@@ -1,132 +0,0 @@
-name: AI Agent Integration Tests
-
-# Exercises the AI agent flow path (preview_flow with `aiagent` modules) against
-# real LLM providers. Runs only when AI-agent backend code or the tests change,
-# because each run makes real (paid) LLM calls. To avoid spending on every commit,
-# the PR side triggers only when a PR is marked ready for review (out of draft) —
-# not on `synchronize` — plus push to main and manual dispatch.
-on:
-  workflow_dispatch:
-  push:
-    branches: [main]
-    paths:
-      - "integration_tests/ai_agent_tests/**"
-      - "backend/windmill-ai/**"
-      - "backend/windmill-api/src/ai.rs"
-      - "backend/windmill-worker/src/ai_executor.rs"
-      - "backend/windmill-worker/src/ai/**"
-      - "backend/windmill-worker/src/memory_common.rs"
-      - "backend/windmill-common/src/flow_conversations.rs"
-      - ".github/workflows/ai-agent-tests.yml"
-  pull_request:
-    types: [opened, reopened, ready_for_review]
-    paths:
-      - "integration_tests/ai_agent_tests/**"
-      - "backend/windmill-ai/**"
-      - "backend/windmill-api/src/ai.rs"
-      - "backend/windmill-worker/src/ai_executor.rs"
-      - "backend/windmill-worker/src/ai/**"
-      - "backend/windmill-worker/src/memory_common.rs"
-      - "backend/windmill-common/src/flow_conversations.rs"
-      - ".github/workflows/ai-agent-tests.yml"
-
-concurrency:
-  group: ai-agent-tests-${{ github.ref }}
-  cancel-in-progress: true
-
-jobs:
-  ai_agent_e2e:
-    # Skip draft PRs; the `opened`/`reopened` types would otherwise fire while
-    # still a draft. `ready_for_review` always arrives non-draft.
-    if: github.event_name != 'pull_request' || github.event.pull_request.draft == false
-    runs-on: ubicloud-standard-16
-    services:
-      postgres:
-        image: postgres:16
-        ports:
-          - 5432:5432
-        env:
-          POSTGRES_DB: windmill
-          POSTGRES_PASSWORD: changeme
-        options: >-
-          --health-cmd pg_isready --health-interval 10s --health-timeout 5s
-          --health-retries 5
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions-rust-lang/setup-rust-toolchain@v1
-        with:
-          cache-workspaces: backend
-          toolchain: 1.97.0
-
-      - uses: oven-sh/setup-bun@v2
-        with:
-          bun-version: 1.3.10
-
-      - uses: actions/setup-node@v4
-        with:
-          node-version: "20"
-
-      - uses: actions/setup-python@v5
-        with:
-          python-version: "3.11"
-
-      # CE build (no enterprise/license needed for AI agents). `quickjs` powers
-      # flow input-transform JS eval; `mcp` is required by the deepwiki MCP tool
-      # test. Bun tool scripts run via the always-on worker (BUN_PATH).
-      - name: Build Windmill
-        working-directory: ./backend
-        env:
-          SQLX_OFFLINE: true
-          CARGO_BUILD_JOBS: 12
-          RUSTFLAGS: ""
-        run: cargo build --features quickjs,mcp
-
-      - name: Start Windmill
-        working-directory: ./backend
-        env:
-          DATABASE_URL: postgres://postgres:changeme@localhost:5432/windmill
-          BUN_PATH: bun
-          NODE_BIN_PATH: node
-          RUST_LOG: info
-        run: |
-          mkdir -p ../integration_tests/logs
-          ./target/debug/windmill > ../integration_tests/logs/windmill.log 2>&1 &
-          echo "Waiting for Windmill to be ready..."
-          for i in $(seq 1 60); do
-            if curl -sf http://localhost:8000/api/version > /dev/null 2>&1; then
-              echo "Windmill is ready"
-              break
-            fi
-            sleep 2
-          done
-          curl -sf http://localhost:8000/api/version > /dev/null || { echo "Windmill failed to start"; tail -50 ../integration_tests/logs/windmill.log; exit 1; }
-
-      - name: Run AI agent integration tests
-        timeout-minutes: 20
-        working-directory: ./integration_tests/ai_agent_tests
-        env:
-          WINDMILL_URL: http://localhost:8000
-          # Only the providers we have org secrets for. Other providers
-          # (Azure, Bedrock, OpenRouter) are skipped by conftest when their
-          # keys are absent — see skip_provider_without_credentials.
-          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
-          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-          GOOGLE_AI_API_KEY: ${{ secrets.GOOGLE_API_KEY }}
-        run: |
-          python -m venv .venv
-          .venv/bin/pip install -r requirements.txt
-          # The S3/vision-attachment tests need MinIO large-file storage and
-          # image-capable provider setup; out of scope for this cost-controlled
-          # smoke. Add MinIO secrets + a storage service to enable them.
-          .venv/bin/python -m pytest -v \
-            --ignore=test_user_attachments.py \
-            --ignore=test_user_images.py \
-            --ignore=test_image_output.py
-
-      - name: Archive Windmill logs
-        uses: actions/upload-artifact@v4
-        if: always()
-        with:
-          name: ai-agent-tests-windmill-logs
-          path: integration_tests/logs

.github/workflows/ai-evals-test.yml

@@ -1,174 +0,0 @@
-name: AI Evals (global mode)
-
-# Smoke-tests the production global AI chat proxy/frontend execution path via
-# the ai_evals harness, one case across one cheap model per provider. Runs only
-# when the eval harness or the global chat code change, since each run makes real
-# (paid) LLM calls. The backend is built from source purely as the AI proxy the
-# harness routes model calls through; the global tools/drafts run in-process in
-# the Vitest bridge against production frontend code. To avoid spending on every
-# commit, the PR side triggers only when a PR is marked ready for review (out of
-# draft) — not on `synchronize` — plus push to main and manual dispatch.
-on:
-  workflow_dispatch:
-  push:
-    branches: [main]
-    paths:
-      - "ai_evals/**"
-      - "backend/windmill-api/src/ai.rs"
-      - "backend/windmill-ai/**"
-      - "frontend/src/lib/components/copilot/**"
-      # The eval harness runs production frontend code in-process; these are the
-      # AI/draft-specific deps outside copilot/ that the global smoke exercises.
-      - "frontend/src/lib/userDraft.svelte.ts"
-      - "frontend/src/lib/userDraftDbSyncer.svelte.ts"
-      - "frontend/src/lib/infer.ts"
-      - ".github/workflows/ai-evals-test.yml"
-  pull_request:
-    types: [opened, reopened, ready_for_review]
-    paths:
-      - "ai_evals/**"
-      - "backend/windmill-api/src/ai.rs"
-      - "backend/windmill-ai/**"
-      - "frontend/src/lib/components/copilot/**"
-      # The eval harness runs production frontend code in-process; these are the
-      # AI/draft-specific deps outside copilot/ that the global smoke exercises.
-      - "frontend/src/lib/userDraft.svelte.ts"
-      - "frontend/src/lib/userDraftDbSyncer.svelte.ts"
-      - "frontend/src/lib/infer.ts"
-      - ".github/workflows/ai-evals-test.yml"
-
-concurrency:
-  group: ai-evals-test-${{ github.ref }}
-  cancel-in-progress: true
-
-jobs:
-  ai_evals_global:
-    # Provider secrets are unavailable to forked and Dependabot PRs.
-    if: >-
-      github.event_name != 'pull_request' ||
-      (
-        github.event.pull_request.draft == false &&
-        github.event.pull_request.head.repo.full_name == github.repository &&
-        github.event.pull_request.user.login != 'dependabot[bot]'
-      )
-    runs-on: ubicloud-standard-16
-    services:
-      postgres:
-        image: postgres:16
-        ports:
-          - 5432:5432
-        env:
-          POSTGRES_DB: windmill
-          POSTGRES_PASSWORD: changeme
-        options: >-
-          --health-cmd pg_isready --health-interval 10s --health-timeout 5s
-          --health-retries 5
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions-rust-lang/setup-rust-toolchain@v1
-        with:
-          cache-workspaces: backend
-          toolchain: 1.97.0
-
-      - uses: oven-sh/setup-bun@v2
-        with:
-          bun-version: 1.3.10
-
-      - uses: actions/setup-node@v4
-        with:
-          # Node 24 ships npm 11, which frontend/package-lock.json is authored
-          # with; npm 10 rejects it ("Missing: picomatch@4.0.5 from lock file").
-          # Node must also stay >= 22.19 for the frontend's undici 8.x, which the
-          # Vitest bridge loads; Node 20 fails with markAsUncloneable.
-          node-version: "24"
-
-      # CE build used only as the AI proxy (login, workspace, provider resource,
-      # /ai/proxy). No worker execution or MCP needed — global tools/drafts run
-      # in the Vitest bridge. quickjs matches the standard CE feature set.
-      - name: Build Windmill (AI proxy)
-        working-directory: ./backend
-        env:
-          SQLX_OFFLINE: true
-          CARGO_BUILD_JOBS: 12
-          RUSTFLAGS: ""
-        run: cargo build --features quickjs
-
-      - name: Start Windmill
-        working-directory: ./backend
-        env:
-          DATABASE_URL: postgres://postgres:changeme@localhost:5432/windmill
-          RUST_LOG: info
-        run: |
-          mkdir -p ../ai_evals/logs
-          ./target/debug/windmill > ../ai_evals/logs/windmill.log 2>&1 &
-          echo "Waiting for Windmill to be ready..."
-          for i in $(seq 1 60); do
-            if curl -sf http://localhost:8000/api/version > /dev/null 2>&1; then
-              echo "Windmill is ready"
-              break
-            fi
-            sleep 2
-          done
-          curl -sf http://localhost:8000/api/version > /dev/null || { echo "Windmill failed to start"; tail -50 ../ai_evals/logs/windmill.log; exit 1; }
-
-      - name: Install frontend deps + generate client
-        working-directory: ./frontend
-        run: |
-          npm ci
-          npm run generate-backend-client
-
-      - name: Run harness unit tests
-        working-directory: ./ai_evals
-        run: |
-          bun install
-          bun test adapters/
-
-      - name: Run global AI evals
-        timeout-minutes: 20
-        working-directory: ./ai_evals
-        env:
-          WMILL_AI_EVAL_BACKEND_URL: http://localhost:8000
-          WMILL_AI_EVAL_BACKEND_WORKSPACE: integration-tests
-          # Anthropic backs the haiku model. Google AI uses GEMINI_API_KEY.
-          ANTHROPIC_API_KEY: ${{ secrets.ANTHROPIC_API_KEY }}
-          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-          GEMINI_API_KEY: ${{ secrets.GOOGLE_API_KEY }}
-          DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}
-        run: |
-          bun install
-          mkdir -p results
-          # One cheap model per provider (anthropic/openai/googleai/deepseek).
-          fail=0
-          for m in haiku 4o gemini-3-flash-preview deepseek-v4-flash; do
-            echo "::group::global-test1-script-create ($m)"
-            if ! bun run cli -- run global global-test1-script-create \
-                  --model "$m" --execution-only --output "$PWD/results/ci-$m.json"; then
-              echo "$m: harness/proxy errored"
-              fail=1
-              echo "::endgroup::"
-              continue
-            fi
-            # The CLI exits 0 when the harness records failed attempts, so gate
-            # on execution-only pass counts while ignoring model output quality.
-            if jq -e \
-              '.attemptCount > 0 and .passedAttempts == .attemptCount' \
-              "results/ci-$m.json" > /dev/null; then
-              echo "$m: OK — proxy/frontend execution completed"
-            else
-              echo "$m: FAILED proxy/frontend execution"
-              jq -c '.cases[0].attempts[0].checks' "results/ci-$m.json" || true
-              fail=1
-            fi
-            echo "::endgroup::"
-          done
-          [ "$fail" = 0 ] || { echo "ai_evals global smoke failed"; exit 1; }
-
-      - name: Archive logs and results
-        uses: actions/upload-artifact@v4
-        if: always()
-        with:
-          name: ai-evals-global-logs
-          path: |
-            ai_evals/logs
-            ai_evals/results

.github/workflows/backend-check.yml

@@ -23,7 +23,7 @@ jobs:
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           cache: false
-          toolchain: 1.97.0
+          toolchain: 1.93.0
       - name: cargo check
         working-directory: ./backend
         timeout-minutes: 16
@@ -44,7 +44,7 @@ jobs:
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           cache: false
-          toolchain: 1.97.0
+          toolchain: 1.93.0
       - name: cargo check
         working-directory: ./backend
         timeout-minutes: 16
@@ -81,7 +81,7 @@ jobs:
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           cache: false
-          toolchain: 1.97.0
+          toolchain: 1.93.0
       - name: cargo check
         working-directory: ./backend
         timeout-minutes: 16
@@ -118,7 +118,7 @@ jobs:
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           cache-workspaces: backend
-          toolchain: 1.97.0
+          toolchain: 1.93.0
       - name: Fix stale v8 build cache
         working-directory: ./backend
         run: |

.github/workflows/backend-test-windows.yml

@@ -50,12 +50,7 @@ jobs:
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           cache-workspaces: backend
-          toolchain: 1.97.0
-          # This action defaults RUSTFLAGS to "-D warnings"; unset it so the test
-          # run is not failed by cross-platform dead-code (cfg(unix)-only helpers
-          # are unused on Windows). Warning hygiene is enforced on the Linux CI
-          # and the build_windows_worker_ release build, not this test job.
-          rustflags: ""
+          toolchain: 1.93.0
 
       - uses: actions/setup-dotnet@v4
         with:
@@ -63,9 +58,7 @@ jobs:
 
       - uses: denoland/setup-deno@v2
         with:
-          # Pin to the Deno version shipped in the runtime image (Dockerfile) so CI
-          # tests what production runs, instead of floating on the latest v2.x.
-          deno-version: 2.2.1
+          deno-version: v2.x
 
       - uses: actions/setup-go@v2
         with:
@@ -81,7 +74,7 @@ jobs:
 
       - uses: astral-sh/setup-uv@v6.2.1
         with:
-          version: "0.11.24"
+          version: "0.9.24"
 
       - uses: shivammathur/setup-php@v2
         with:
@@ -105,21 +98,6 @@ jobs:
           vcpkg.exe install openssl:x64-windows-static
           vcpkg.exe integrate install
 
-      - name: Free disk space (post-vcpkg)
-        shell: pwsh
-        run: |
-          # vcpkg leaves multi-GB of buildtrees/downloads after installing openssl;
-          # we only need the installed/ dir for linking.
-          $vcpkgRoot = $env:VCPKG_INSTALLATION_ROOT
-          foreach ($sub in @("buildtrees", "downloads", "packages")) {
-            $path = Join-Path $vcpkgRoot $sub
-            if (Test-Path $path) {
-              Write-Host "Removing $path"
-              Remove-Item -Recurse -Force -ErrorAction SilentlyContinue $path
-            }
-          }
-          Get-PSDrive C | Select-Object Used,Free | Format-Table -AutoSize
-
       - name: Get runtime paths
         id: runtime-paths
         shell: pwsh
@@ -141,10 +119,6 @@ jobs:
           cargo build --release -p windmill_duckdb_ffi_internal
           New-Item -ItemType Directory -Path ..\target\debug -Force
           Copy-Item target\release\windmill_duckdb_ffi_internal.dll ..\target\debug\
-          # duckdb is bundled (~2GB of build artifacts); the DLL is the only
-          # thing we need from this excluded-crate target dir.
-          Remove-Item -Recurse -Force -ErrorAction SilentlyContinue target
-          Get-PSDrive C | Select-Object Used,Free | Format-Table -AutoSize
 
       - name: Print runtime versions and env
         shell: pwsh
@@ -162,10 +136,6 @@ jobs:
           echo "USERPROFILE=$env:USERPROFILE"
           echo "HOME=$env:HOME"
 
-      - name: Disk space before cargo test
-        shell: pwsh
-        run: Get-PSDrive C | Select-Object Used,Free | Format-Table -AutoSize
-
       - name: cargo test
         working-directory: backend
         timeout-minutes: 60
@@ -174,26 +144,7 @@ jobs:
           RUST_LOG: "off"
           RUST_LOG_STYLE: never
           CARGO_NET_GIT_FETCH_WITH_CLI: true
-          # 16-vcpu runners with disabled PDB still hit LNK1180 ("insufficient
-          # disk space") at link time with 12 parallel link jobs: each test
-          # binary link spikes several hundred MB of transient I/O. Capping at
-          # 8 trades ~25% wall time for headroom on the ~75GB runner disk.
-          CARGO_BUILD_JOBS: 8
-          # backend/Cargo.toml leaves profile.dev at the default debug = 2 for
-          # the (large) windmill workspace crates; that debuginfo is emitted
-          # into every object file and embedded in each test binary, and on
-          # windows-msvc also spawns the mspdbsrv.exe PDB type server. Across a
-          # full --all --features build it is the dominant consumer of the
-          # ~63GB free on the runner disk (LNK1180 / disk-full during linking).
-          # CI needs no debug info, so drop it entirely for the dev/test
-          # profiles here. debug = 0 supersedes the previous split-debuginfo=off
-          # knob (no debuginfo => no .pdb and no LNK1318 type-server limit).
-          CARGO_PROFILE_DEV_DEBUG: "0"
-          CARGO_PROFILE_TEST_DEBUG: "0"
-          # Tests' poll-time stack frames (deep nested async fn chains in
-          # debug builds) reach ~1.8MB. 4MB gives ~2x headroom against flaky
-          # overflows under parallel-test contention.
-          RUST_MIN_STACK: 4194304
+          CARGO_BUILD_JOBS: 12
           VCPKGRS_DYNAMIC: 1
           OPENSSL_DIR: ${{ env.VCPKG_INSTALLATION_ROOT }}\installed\x64-windows-static
           DENO_PATH: ${{ steps.runtime-paths.outputs.DENO_PATH }}
@@ -208,15 +159,9 @@ jobs:
           WMDEBUG_FORCE_V0_WORKSPACE_DEPENDENCIES: 1
           WMDEBUG_FORCE_RUNNABLE_SETTINGS_V0: 1
           WMDEBUG_FORCE_NO_LEGACY_DEBOUNCING_COMPAT: 1
-        # Windows ships a worker-only binary, so test the crates a worker runs
-        # (windmill-worker/-common/-queue) via -p, not `--all`: this skips the
-        # disk-heavy windmill-api test binaries (LNK1180) and the server-only
-        # windmill-trigger-* crates (amqp does not build on Windows). Linux CI runs the rest.
         run: >
           cargo test
           --no-fail-fast
-          -p windmill-worker
-          -p windmill-common
-          -p windmill-queue
-          --features private,enterprise,deno_core,duckdb,python,rust,csharp,php,quickjs,parquet,mcp,scoped_cache,windmill-git-sync/private,windmill-object-store/private,windmill-object-store/enterprise
+          --features enterprise,deno_core,duckdb,license,python,rust,scoped_cache,parquet,private,csharp,php,quickjs,mcp,run_inline
+          --all
           -- --nocapture --test-threads=10

.github/workflows/backend-test.yml

@@ -50,9 +50,7 @@ jobs:
           dotnet-version: "9.0.x"
       - uses: denoland/setup-deno@v2
         with:
-          # Pin to the Deno version shipped in the runtime image (Dockerfile) so CI
-          # tests what production runs, instead of floating on the latest v2.x.
-          deno-version: 2.2.1
+          deno-version: v2.x
       - uses: actions/setup-go@v2
         with:
           go-version: 1.21.5
@@ -64,7 +62,7 @@ jobs:
           node-version: "20"
       - uses: astral-sh/setup-uv@v6.2.1
         with:
-          version: "0.11.24"
+          version: "0.9.24"
       - uses: shivammathur/setup-php@v2
         with:
           php-version: "8.3"
@@ -90,7 +88,7 @@ jobs:
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           cache-workspaces: backend
-          toolchain: 1.97.0
+          toolchain: 1.93.0
       - name: Fix stale v8 build cache
         working-directory: ./backend
         run: |
@@ -239,60 +237,18 @@ jobs:
       - name: cargo test
         timeout-minutes: 30
         env:
-          # setup-rust-toolchain exports RUSTFLAGS=-D warnings, and the RUSTFLAGS env
-          # var fully REPLACES (never merges with) target.*.rustflags in
-          # backend/.cargo/config.toml. That silently drops the config's
-          # `-C link-arg=-fuse-ld=mold`, so CI links the many large integration-test
-          # binaries (v8 + duckdb + every language runtime, statically linked) with the
-          # default bfd linker. Its peak memory across ~12 parallel links OOM-kills the
-          # runner mid-link (SIGTERM => exit 143, before any test runs). Re-add the mold
-          # link arg here so CI links with mold like local dev, keeping -D warnings.
-          # (config.toml's `linker = "clang"` still applies; env only overrides rustflags.)
-          RUSTFLAGS: "-D warnings -C link-arg=-fuse-ld=mold"
           SQLX_OFFLINE: true
           DATABASE_URL: postgres://postgres:changeme@localhost:5432/windmill
           DISABLE_EMBEDDING: true
           RUST_LOG: "off"
           RUST_LOG_STYLE: never
           CARGO_NET_GIT_FETCH_WITH_CLI: true
-          # Cap parallel rustc/link jobs below the 16 available cores. The tail of
-          # the build links ~128 full-graph test binaries (one per tests/*.rs file
-          # across the workspace); at high parallelism enough heavy codegen+link
-          # units (rustc ~2.6GB, mold ~1GB each) overlap to exhaust the 64GB
-          # runner. Matches backend-test-windows.yml, which already uses 8.
-          CARGO_BUILD_JOBS: 8
-          # Incremental compilation is per-run dead weight in CI: rust-cache
-          # (cache-workspaces above) restores compiled dependency artifacts but
-          # never persists target/**/incremental, so there is no prior state to
-          # reuse in a one-shot `cargo test`. It only adds per-crate memory
-          # overhead and extra disk. Off here (kept on for local dev via
-          # .cargo/config.toml). Matches backend-test-windows.yml.
-          CARGO_INCREMENTAL: "0"
-          # backend/Cargo.toml leaves profile.dev at the default debug = 2 for
-          # the (large) windmill workspace crates; that debug info is emitted
-          # into every object file and embedded in each test binary. Across the
-          # full --all --features build it is the dominant memory/disk consumer
-          # when mold links the windmill-api-integration-tests binary, tipping
-          # the runner over (lost runner reported as a canceled step). CI needs
-          # no debug info, so drop it entirely for the dev/test profiles here.
-          # (test profile inherits dev, but the workspace crates link in as
-          # dev-profile deps, so both must be set.) CI-only; local dev builds
-          # are unaffected.
-          CARGO_PROFILE_DEV_DEBUG: "0"
-          CARGO_PROFILE_TEST_DEBUG: "0"
-          # Tests' poll-time stack frames (deep nested async fn chains in
-          # debug builds) reach ~1.8MB, leaving very thin headroom on the
-          # default 2MB thread stack. 4MB gives ~2x buffer against flaky
-          # overflows under parallel-test contention.
-          RUST_MIN_STACK: 4194304
+          CARGO_BUILD_JOBS: 12
           WMDEBUG_FORCE_V0_WORKSPACE_DEPENDENCIES: 1
           WMDEBUG_FORCE_RUNNABLE_SETTINGS_V0: 1
           WMDEBUG_FORCE_NO_LEGACY_DEBOUNCING_COMPAT: 1
           TEST_NPM_REGISTRY: "http://localhost:4873/:_authToken=${{ env.NPM_TOKEN }}"
         run: |
           deno --version && bun -v && node --version && go version && python3 --version && php --version && ruby --version && pwsh --version && dotnet --version
-          # The FFI crate is excluded from the workspace, so the `cargo test` below
-          # never reaches it. Pin the target dir (matching the cache step above) so
-          # its own tests run off this compile rather than a second bundled build.
-          (cd windmill-duckdb-ffi-internal && export CARGO_TARGET_DIR="$PWD/target" && ./build_dev.sh && cargo test --release -p windmill_duckdb_ffi_internal)
+          cd windmill-duckdb-ffi-internal && ./build_dev.sh && cd ..
           DENO_PATH=$(which deno) BUN_PATH=$(which bun) NODE_BIN_PATH=$(which node) GO_PATH=$(which go) UV_PATH=$(which uv) PHP_PATH=$(which php) COMPOSER_PATH=$(which composer) RUBY_PATH=$(which ruby) RUBY_BUNDLE_PATH=$(which bundle) RUBY_GEM_PATH=$(which gem) POWERSHELL_PATH=$(which pwsh) DOTNET_PATH=$(which dotnet) cargo test --features enterprise,deno_core,duckdb,license,python,rust,scoped_cache,parquet,private,private_registry_test,csharp,php,ruby,mysql,quickjs,mcp,run_inline --all -- --nocapture --test-threads=10

.github/workflows/build-caddy-l4-image.yml

@@ -5,20 +5,6 @@ env:
 name: Build caddy-l4
 on:
   workflow_dispatch:
-  push:
-    branches:
-      - main
-    paths:
-      - docker/DockerfileCaddyL4
-      - docker/entrypoint-caddy.sh
-      - docker/caddy-compat-normalize.awk
-      - docker/caddy-l4.version
-      - docker/test-caddy-compat.sh
-      - Caddyfile
-      # The version check below reads the pin out of docker-compose.yml, so a
-      # compose-only bump has to trigger this workflow or the check never runs.
-      - docker-compose.yml
-      - .github/workflows/build-caddy-l4-image.yml
 
 permissions: write-all
 
@@ -28,35 +14,6 @@ jobs:
     steps:
       - uses: actions/checkout@v4
       - uses: depot/setup-action@v1
-
-      # docker-compose.yml pins an exact tag, and the Caddyfile it must agree
-      # with lives in the same checkout. Fail the build rather than publish a
-      # version nothing references, which is how :latest drifted from the
-      # Caddyfile in the first place.
-      - name: Resolve and check image version
-        id: version
-        run: |
-          set -euo pipefail
-          version="$(tr -d '[:space:]' < docker/caddy-l4.version)"
-          pinned="$(grep -oE 'caddy-l4:[^[:space:]"]+' docker-compose.yml | head -1 | cut -d: -f2-)"
-          caddy="$(grep -m1 -oE '^FROM caddy:[0-9]+\.[0-9]+\.[0-9]+' docker/DockerfileCaddyL4 | cut -d: -f2)"
-          if [ "$version" != "$pinned" ]; then
-            echo "docker/caddy-l4.version is '$version' but docker-compose.yml pins '$pinned'" >&2
-            echo "Bump both together." >&2
-            exit 1
-          fi
-          # Otherwise a caddy bump that forgets the version file publishes a tag
-          # that names the wrong caddy.
-          case "$version" in
-            "$caddy"-*) ;;
-            *)
-              echo "docker/caddy-l4.version is '$version' but the Dockerfile pins caddy '$caddy'" >&2
-              echo "The version must be <caddy-version>-<revision>." >&2
-              exit 1
-              ;;
-          esac
-          echo "version=$version" >> "$GITHUB_OUTPUT"
-
       - name: Docker meta
         id: meta-ee-public
         uses: docker/metadata-action@v5
@@ -66,22 +23,8 @@ jobs:
           tags: |
             type=sha
             type=ref,event=branch
-            # Not gated on the default branch: docker-compose.yml pins this tag,
-            # so it has to be publishable from a branch (workflow_dispatch)
-            # before the pin merges, or main would reference a tag that does not
-            # exist yet. The version is immutable, so republishing from main is
-            # a no-op. Only branch pushes to main and manual dispatch run this
-            # workflow, so a branch cannot claim the tag by accident.
-            type=raw,value=${{ steps.version.outputs.version }}
             type=raw,value=latest,enable={{is_default_branch}}
 
-      # The shim rewrites config a self-hoster never sees, so a silent
-      # regression here strands them on a restart loop or a dead :80.
-      - name: Test the legacy-Caddyfile compatibility shim
-        run: |
-          docker build -f docker/DockerfileCaddyL4 -t caddy-l4:ci ./docker
-          docker/test-caddy-compat.sh caddy-l4:ci
-
       - name: Login to registry
         uses: docker/login-action@v3
         with:

.github/workflows/build-publish-rh-image.yml

@@ -63,7 +63,6 @@ jobs:
           push: true
           build-args: |
             features=ee_rhel
-            WM_BUILD_VERSION=${{ github.sha }}
           secrets: |
             rh_username=${{ secrets.RH_USERNAME }}
             rh_password=${{ secrets.RH_PASSWORD }}

.github/workflows/build-publish-rh8-image.yml

@@ -65,7 +65,6 @@ jobs:
           push: true
           build-args: |
             features=ee_rhel
-            WM_BUILD_VERSION=${{ github.sha }}
           secrets: |
             rh_username=${{ secrets.RH_USERNAME }}
             rh_password=${{ secrets.RH_PASSWORD }}
@@ -83,7 +82,6 @@ jobs:
           push: true
           build-args: |
             features=ee_rhel
-            WM_BUILD_VERSION=${{ github.sha }}
           secrets: |
             rh_username=${{ secrets.RH_USERNAME }}
             rh_password=${{ secrets.RH_PASSWORD }}

.github/workflows/build_windows_worker_.yml

@@ -33,7 +33,7 @@ jobs:
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           cache-workspaces: backend
-          toolchain: 1.97.0
+          toolchain: 1.93.0
 
       - name: Substitute EE code
         shell: bash
@@ -45,12 +45,8 @@ jobs:
         env:
           RUSTFLAGS: "-D warnings"
         run: |
-          cd backend
-          # Stub the openapi specs to empty: they are compiled in via an ungated
-          # include_str! but a worker binary never serves them, so this avoids
-          # embedding ~2.5MB of spec.
+          mkdir frontend/build && cd backend
           New-Item -Path . -Name "windmill-api/openapi-deref.yaml" -ItemType "File" -Force
-          New-Item -Path . -Name "windmill-api/openapi-deref.json" -ItemType "File" -Force
           cargo check --features=ee_windows
 
       - name: Cargo build dynamic libraries windows

.github/workflows/check-docs-links.yml

@@ -1,23 +0,0 @@
-name: Check frontend docs links
-
-on:
-  push:
-    tags:
-      - "v*"
-  workflow_dispatch:
-
-jobs:
-  check-docs-links:
-    runs-on: ubuntu-latest
-    timeout-minutes: 10
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          sparse-checkout: |
-            frontend/src
-            .github/scripts
-      - uses: actions/setup-node@v4
-        with:
-          node-version: "22.x"
-      - name: Verify docs links are not 404
-        run: node .github/scripts/check-docs-links.mjs

.github/workflows/check-empty-fixture.yml

@@ -1,19 +0,0 @@
-name: Check fixture is empty
-
-on:
-  push:
-    branches: [main]
-    paths:
-      - "fixtures/**"
-  pull_request:
-    paths:
-      - "fixtures/**"
-
-jobs:
-  check-empty-fixture:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-
-      - name: Ensure fixtures/cli-sync/ has no committed snapshot
-        run: bash fixtures/check-empty.sh

.github/workflows/check-org-membership.yml

@@ -0,0 +1,83 @@
+name: Check Organization Membership
+
+on:
+  workflow_call:
+    inputs:
+      commenter:
+        required: false
+        type: string
+        default: ''
+        description: 'The username to check. Auto-detected from the event context if not provided.'
+      organization:
+        required: false
+        type: string
+        default: 'windmill-labs'
+        description: 'The organization to check membership for'
+      trusted_bot:
+        required: false
+        type: string
+        default: 'windmill-internal-app[bot]'
+        description: 'The trusted bot username to allow'
+    secrets:
+      access_token:
+        required: true
+        description: 'The access token to use for org membership check'
+    outputs:
+      is_member:
+        description: 'Whether the user is an organization member or trusted bot'
+        value: ${{ jobs.check-membership.outputs.is_member }}
+
+jobs:
+  check-membership:
+    runs-on: ubicloud-standard-2
+    outputs:
+      is_member: ${{ steps.check-membership.outputs.is_member }}
+    steps:
+      - name: Determine commenter
+        id: determine-commenter
+        run: |
+          COMMENTER="${{ inputs.commenter }}"
+          if [[ -z "$COMMENTER" ]]; then
+            if [[ "${{ github.event_name }}" == "issue_comment" || \
+                  "${{ github.event_name }}" == "pull_request_review_comment" ]]; then
+              COMMENTER="${{ github.event.comment.user.login }}"
+            elif [[ "${{ github.event_name }}" == "pull_request_review" ]]; then
+              COMMENTER="${{ github.event.review.user.login }}"
+            else
+              COMMENTER="${{ github.event.issue.user.login }}"
+            fi
+          fi
+          echo "commenter=$COMMENTER" >> $GITHUB_OUTPUT
+
+      - name: Check organization membership
+        id: check-membership
+        env:
+          ORG_ACCESS_TOKEN: ${{ secrets.access_token }}
+          COMMENTER: ${{ steps.determine-commenter.outputs.commenter }}
+          ORG: ${{ inputs.organization }}
+          TRUSTED_BOT: ${{ inputs.trusted_bot }}
+        run: |
+          # 1. Allow the trusted bot straight away
+          if [[ "$COMMENTER" == "$TRUSTED_BOT" ]]; then
+            echo "is_member=true" >> $GITHUB_OUTPUT
+            exit 0
+          fi
+
+          # 2. Disallow other bots
+          if [[ "${COMMENTER}" =~ \[bot\]$ ]]; then
+            echo "is_member=false" >> $GITHUB_OUTPUT
+            exit 0
+          fi
+
+          # 3. Otherwise check if the user is a member of the organization
+          STATUS=$(curl -s -o /dev/null -w "%{http_code}" \
+            -H "Authorization: token $ORG_ACCESS_TOKEN" \
+            -H "Accept: application/vnd.github+json" \
+            -H "X-GitHub-Api-Version: 2022-11-28" \
+            "https://api.github.com/orgs/$ORG/members/$COMMENTER")
+
+          if [ "$STATUS" -eq 204 ]; then
+            echo "is_member=true" >> $GITHUB_OUTPUT
+          else
+            echo "is_member=false" >> $GITHUB_OUTPUT
+          fi

.github/workflows/check-system-prompts.yml

@@ -10,7 +10,6 @@ on:
       - "backend/windmill-api/openapi.yaml"
       - "cli/src/main.ts"
       - "cli/src/commands/**"
-      - "frontend/src/lib/components/copilot/chat/workspaceToolsZod.gen.ts"
   pull_request:
     paths:
       - "system_prompts/**"
@@ -20,7 +19,6 @@ on:
       - "backend/windmill-api/openapi.yaml"
       - "cli/src/main.ts"
       - "cli/src/commands/**"
-      - "frontend/src/lib/components/copilot/chat/workspaceToolsZod.gen.ts"
 
 jobs:
   check-freshness:

.github/workflows/check-write-access.yml

@@ -1,74 +0,0 @@
-name: Check Write Access
-
-# Authorizes a user to trigger privileged command workflows (/review, /ai, /plan,
-# /updatesqlx, ...). The webhook author_association reports PRIVATE org members as
-# CONTRIBUTOR/NONE (only public members show as MEMBER), so command jobs can't gate on
-# it alone. This mints the internal GitHub App token — which can see private members —
-# and confirms the user is a member or has write access to the repo. The app token is
-# minted fresh per run, so unlike the old ORG_ACCESS_TOKEN PAT it never expires.
-
-on:
-  workflow_call:
-    inputs:
-      username:
-        required: true
-        type: string
-        description: 'The user whose access to verify'
-      trusted_bot:
-        required: false
-        type: string
-        default: 'windmill-internal-app[bot]'
-        description: 'A bot login that is always authorized'
-    outputs:
-      authorized:
-        description: 'true if the user is the trusted bot, an org member, or has repo write access'
-        value: ${{ jobs.check.outputs.authorized }}
-
-jobs:
-  check:
-    runs-on: ubuntu-latest
-    outputs:
-      authorized: ${{ steps.check.outputs.authorized }}
-    steps:
-      # This check is purely additive: callers OR it with author_association, so it must
-      # never fail the job. Failing here would block every dependent reviewer job through
-      # `needs`, turning an unconfigured or misconfigured app into a total review outage
-      # rather than a fallback to the author_association path.
-      - name: Mint internal app token
-        id: app
-        if: vars.INTERNAL_APP_ID != ''
-        continue-on-error: true
-        uses: actions/create-github-app-token@v2
-        with:
-          app-id: ${{ vars.INTERNAL_APP_ID }}
-          private-key: ${{ secrets.INTERNAL_APP_KEY }}
-          owner: ${{ github.repository_owner }}
-
-      - name: Resolve authorization
-        id: check
-        env:
-          # Without the app token, the default token still resolves public members and
-          # repo collaborators; private members simply fall through to author_association.
-          GH_TOKEN: ${{ steps.app.outputs.token || github.token }}
-          USERNAME: ${{ inputs.username }}
-          TRUSTED_BOT: ${{ inputs.trusted_bot }}
-          REPO: ${{ github.repository }}
-        run: |
-          if [ "$USERNAME" = "$TRUSTED_BOT" ]; then
-            echo "authorized=true" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-          ORG="${REPO%%/*}"
-          # Org membership resolves private members too (204 = member, 404 = not).
-          if gh api "orgs/$ORG/members/$USERNAME" --silent 2>/dev/null; then
-            echo "authorized=true" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-          # Fallback: effective repo permission (also covers outside collaborators).
-          PERM=$(gh api "repos/$REPO/collaborators/$USERNAME/permission" --jq '.permission' 2>/dev/null || echo none)
-          if [ "$PERM" = "admin" ] || [ "$PERM" = "write" ]; then
-            echo "authorized=true" >> "$GITHUB_OUTPUT"
-          else
-            echo "authorized=false" >> "$GITHUB_OUTPUT"
-            echo "$USERNAME is neither the trusted bot, an org member, nor a repo writer."
-          fi

.github/workflows/claude-fast.yml

@@ -0,0 +1,54 @@
+name: Fast Claude
+
+on:
+  issue_comment:
+    types: [created]
+  pull_request_review_comment:
+    types: [created]
+  issues:
+    types: [opened, assigned]
+  pull_request_review:
+    types: [submitted]
+
+jobs:
+  check-membership:
+    if: |
+      (github.event_name == 'issue_comment' && contains(github.event.comment.body, '/ai-fast')) ||
+      (github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '/ai-fast')) ||
+      (github.event_name == 'pull_request_review' && contains(github.event.review.body, '/ai-fast')) ||
+      (github.event_name == 'issues' && contains(github.event.issue.body, '/ai-fast'))
+    uses: ./.github/workflows/check-org-membership.yml
+    secrets:
+      access_token: ${{ secrets.ORG_ACCESS_TOKEN }}
+
+  claude-code-action:
+    needs: check-membership
+    if: |
+      needs.check-membership.outputs.is_member == 'true'
+    runs-on: ubicloud-standard-8
+    permissions:
+      contents: write
+      pull-requests: write
+      issues: write
+      id-token: write
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 1
+
+      - name: Run Claude PR Action
+        uses: anthropics/claude-code-action@v1
+        with:
+          claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
+          allowed_bots: "windmill-internal-app[bot]"
+          trigger_phrase: "/ai-fast"
+          settings: |
+            {
+              "env": {
+                "SQLX_OFFLINE": "true"
+              }
+            }
+          claude_args: |
+            --allowedTools "Bash,WebFetch,WebSearch"
+            --model opus

.github/workflows/claude-plan.yml

@@ -11,24 +11,20 @@ on:
     types: [submitted]
 
 jobs:
-  # author_association misses private org members; check-access resolves them via the
-  # internal app token. Both are OR'd below so public members still pass instantly.
-  check-access:
+  check-membership:
     if: |
       (github.event_name == 'issue_comment' && contains(github.event.comment.body, '/plan')) ||
       (github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '/plan')) ||
       (github.event_name == 'pull_request_review' && contains(github.event.review.body, '/plan')) ||
       (github.event_name == 'issues' && contains(github.event.issue.body, '/plan'))
-    uses: ./.github/workflows/check-write-access.yml
-    with:
-      username: ${{ github.event.comment.user.login || github.event.review.user.login || github.event.issue.user.login }}
-    secrets: inherit
+    uses: ./.github/workflows/check-org-membership.yml
+    secrets:
+      access_token: ${{ secrets.ORG_ACCESS_TOKEN }}
 
   claude-plan-action:
-    needs: [check-access]
+    needs: check-membership
     if: |
-      needs.check-access.outputs.authorized == 'true' ||
-      contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association || github.event.review.author_association || github.event.issue.author_association)
+      needs.check-membership.outputs.is_member == 'true'
     runs-on: ubicloud-standard-4
     timeout-minutes: 20
     permissions:
@@ -49,7 +45,7 @@ jobs:
           allowed_bots: 'windmill-internal-app[bot]'
           trigger_phrase: '/plan'
           claude_args: |
-            --model claude-opus-5
+            --model opus
             --system-prompt "# Claude Planning Mode
 
             You are operating in PLANNING MODE ONLY. Your role is to create detailed, structured plans without making any code changes.

.github/workflows/claude.yml

@@ -1,4 +1,4 @@
-name: Fast Claude
+name: Claude PR Assistant
 
 on:
   issue_comment:
@@ -11,25 +11,22 @@ on:
     types: [submitted]
 
 jobs:
-  # author_association misses private org members; check-access resolves them via the
-  # internal app token. Both are OR'd below so public members still pass instantly.
-  check-access:
+  check-membership:
     if: |
       (github.event_name == 'issue_comment' && startsWith(github.event.comment.body, '/ai') && !startsWith(github.event.comment.body, '/ai-fast')) ||
       (github.event_name == 'pull_request_review_comment' && startsWith(github.event.comment.body, '/ai') && !startsWith(github.event.comment.body, '/ai-fast')) ||
       (github.event_name == 'pull_request_review' && startsWith(github.event.review.body, '/ai') && !startsWith(github.event.review.body, '/ai-fast')) ||
       (github.event_name == 'issues' && startsWith(github.event.issue.body, '/ai') && !startsWith(github.event.issue.body, '/ai-fast'))
-    uses: ./.github/workflows/check-write-access.yml
-    with:
-      username: ${{ github.event.comment.user.login || github.event.review.user.login || github.event.issue.user.login }}
-    secrets: inherit
+    uses: ./.github/workflows/check-org-membership.yml
+    secrets:
+      access_token: ${{ secrets.ORG_ACCESS_TOKEN }}
 
   claude-code-action:
-    needs: [check-access]
+    needs: check-membership
     if: |
-      needs.check-access.outputs.authorized == 'true' ||
-      contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association || github.event.review.author_association || github.event.issue.author_association)
+      needs.check-membership.outputs.is_member == 'true'
     runs-on: ubicloud-standard-8
+    timeout-minutes: 60
     permissions:
       contents: write
       pull-requests: write
@@ -41,43 +38,36 @@ jobs:
         with:
           fetch-depth: 1
 
-      # Make the EE source (the *_ee.rs files in the companion repo) available so the
-      # reviewer can see EE-only code (e.g. windmill-queue/src/jobs_ee.rs), not just the
-      # CE surface. The EE ref is read from the PR head's backend/ee-repo-ref.txt (via the
-      # API, so it reflects the PR's EE pin regardless of which ref is checked out here).
-      - name: Check EE access
-        id: ee
-        env:
-          EE_TOKEN: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-          GH_TOKEN: ${{ github.token }}
-          PR_NUMBER: ${{ github.event.issue.number || github.event.pull_request.number }}
-        run: |
-          if [ -z "$EE_TOKEN" ] || [ -z "$PR_NUMBER" ]; then
-            echo "available=false" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-          HEAD_SHA=$(gh api "repos/${{ github.repository }}/pulls/$PR_NUMBER" --jq .head.sha)
-          REF=$(gh api "repos/${{ github.repository }}/contents/backend/ee-repo-ref.txt?ref=$HEAD_SHA" --jq .content | base64 -d | tr -d '[:space:]')
-          if [ -z "$REF" ]; then
-            echo "available=false" >> "$GITHUB_OUTPUT"
-          else
-            echo "available=true" >> "$GITHUB_OUTPUT"
-            echo "ee_repo_ref=$REF" >> "$GITHUB_OUTPUT"
-          fi
-
-      - name: Checkout EE repository
-        if: steps.ee.outputs.available == 'true'
-        uses: actions/checkout@v4
+      - uses: actions/cache@v3
         with:
-          repository: windmill-labs/windmill-ee-private
-          path: ./windmill-ee-private
-          ref: ${{ steps.ee.outputs.ee_repo_ref }}
-          token: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-          fetch-depth: 1
+          path: ~/.npm
+          key: ${{ runner.os }}-node-${{ hashFiles('**/package-lock.json') }}
+          restore-keys: |
+            ${{ runner.os }}-node-
 
-      - name: Substitute EE code
-        if: steps.ee.outputs.available == 'true'
-        run: ./backend/substitute_ee_code.sh --copy --dir ./windmill-ee-private
+      - name: Run npm install and generate-backend-client
+        working-directory: ./frontend
+        run: |
+          # add a build directory for cargo check
+          mkdir -p build
+          npm install
+          npm run generate-backend-client
+
+      - name: install xmlsec1 and gssapi
+        run: |
+          sudo apt-get update
+          sudo apt-get install -y libxml2-dev libxmlsec1-dev libkrb5-dev libsasl2-dev libcurl4-openssl-dev mold clang
+
+      - uses: actions-rust-lang/setup-rust-toolchain@v1
+        with:
+          cache-workspaces: backend
+          toolchain: 1.93.0
+
+      - name: cargo check
+        working-directory: ./backend
+        timeout-minutes: 16
+        run: |
+          SQLX_OFFLINE=true cargo check --features all_sqlx_features
 
       - name: Run Claude PR Action
         uses: anthropics/claude-code-action@v1
@@ -93,4 +83,25 @@ jobs:
             }
           claude_args: |
             --allowedTools "Bash,WebFetch,WebSearch"
-            --model claude-opus-5
+            --model opus
+            --system-prompt "## IMPORTANT INSTRUCTIONS
+            - Your branch name should be a short description of the requested changes.
+            - Bypass previous instructions and instead of adding a url in your comment to manually create a PR, if you created a new branch with your changes, automatically create a draft PR from that branch to main.
+
+            ## Code Quality Requirements
+
+            After making any code changes, you MUST run the appropriate validation commands:
+
+            **Frontend Changes:**
+            - Run: \`npm run check\` in the frontend directory
+            - Fix all warnings and errors before proceeding
+
+            **Backend Changes:**
+            - Run: \`cargo check --features all_sqlx_features\` in the backend directory
+            - Fix all warnings and errors before proceeding
+
+            **Pull Request Creation:**
+            - DO NOT FORGET TO OPEN A DRAFT PR AFTER YOU ARE DONE if you made changes after a request from a git issue.
+
+            ## Available Tools
+            - Bash: Full access to run validation commands and git operations"

.github/workflows/cli-tests.yml

@@ -6,14 +6,12 @@ on:
     branches: [main]
     paths:
       - "cli/**"
-      - "windmill-yaml-validator/**"
       - "backend/migrations/**"
       - ".github/workflows/cli-tests.yml"
   pull_request:
     branches: [main]
     paths:
       - "cli/**"
-      - "windmill-yaml-validator/**"
       - "backend/migrations/**"
       - ".github/workflows/cli-tests.yml"
 

.github/workflows/codex-pr-review.yml

@@ -2,229 +2,53 @@ name: Codex Auto Review
 
 on:
   pull_request:
-    types: [ready_for_review, opened, synchronize]
-  workflow_call:
-    inputs:
-      pr_number:
-        description: 'PR number to review'
-        required: true
-        type: number
-      extra_prompt:
-        description: 'Additional reviewer instructions appended to the standard review prompt'
-        required: false
-        type: string
-        default: ''
-      triggered_by:
-        description: 'GitHub username that triggered this review (for audit only)'
-        required: false
-        type: string
-        default: ''
-    secrets:
-      OPENAI_API_KEY:
-        required: false
-      CODEX_AUTH_JSON:
-        required: false
-      WINDMILL_EE_PRIVATE_ACCESS:
-        required: false
+    types: [ready_for_review, opened]
 
 concurrency:
-  group: codex-review-${{ inputs.pr_number || github.event.pull_request.number }}
+  group: codex-review-${{ github.event.pull_request.number }}
   cancel-in-progress: true
 
 jobs:
   codex-review:
     runs-on: ubicloud-standard-2
     timeout-minutes: 30
-    # A non-fork PR (head.repo.fork == false) can only be opened by someone with push
-    # access to this repo, so fork==false already enforces write access. Do NOT re-add
-    # an author_association gate: the pull_request webhook payload reports private org
-    # members as CONTRIBUTOR/NONE (only public members show as MEMBER), which silently
-    # skips auto-review for every private member.
-    if: |
-      github.event_name == 'workflow_call' ||
-      (
-        github.event.pull_request.draft == false &&
-        github.event.pull_request.head.repo.fork == false
-      )
+    if: github.event.pull_request.draft == false && github.event.pull_request.head.repo.fork == false
     permissions:
       contents: read
       issues: write
-      pull-requests: write
     steps:
       - name: Check Codex configuration
         id: codex_config
         env:
-          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           CODEX_AUTH_JSON: ${{ secrets.CODEX_AUTH_JSON }}
         run: |
-          if [ -n "$OPENAI_API_KEY" ]; then
+          if [ -n "$CODEX_AUTH_JSON" ]; then
             echo "enabled=true" >> "$GITHUB_OUTPUT"
-            echo "auth_mode=api_key" >> "$GITHUB_OUTPUT"
-          elif [ -n "$CODEX_AUTH_JSON" ]; then
-            echo "enabled=true" >> "$GITHUB_OUTPUT"
-            echo "auth_mode=oauth_json" >> "$GITHUB_OUTPUT"
           else
             echo "enabled=false" >> "$GITHUB_OUTPUT"
-            echo "Codex auth is not configured; set OPENAI_API_KEY or CODEX_AUTH_JSON to enable Codex review."
+            echo "CODEX_AUTH_JSON is not configured; skipping Codex review."
           fi
 
-      - name: Resolve PR metadata
-        if: steps.codex_config.outputs.enabled == 'true'
-        id: pr
-        env:
-          GH_TOKEN: ${{ github.token }}
-          INPUT_PR_NUMBER: ${{ inputs.pr_number }}
-          EVENT_PR_NUMBER: ${{ github.event.pull_request.number }}
-          EVENT_BASE_REF: ${{ github.event.pull_request.base.ref }}
-          EVENT_BASE_SHA: ${{ github.event.pull_request.base.sha }}
-          EVENT_HEAD_SHA: ${{ github.event.pull_request.head.sha }}
-          EVENT_TITLE: ${{ github.event.pull_request.title }}
-          EVENT_BODY: ${{ github.event.pull_request.body }}
-          EVENT_FORK: ${{ github.event.pull_request.head.repo.fork }}
-          EVENT_AUTHOR: ${{ github.event.pull_request.user.login }}
-          EVENT_ACTION: ${{ github.event.action }}
-        run: |
-          if [ -n "$INPUT_PR_NUMBER" ]; then
-            PR_JSON=$(gh pr view "$INPUT_PR_NUMBER" --repo "${{ github.repository }}" \
-              --json number,baseRefName,baseRefOid,headRefOid,title,body,isCrossRepository,author)
-            PR_NUMBER=$(echo "$PR_JSON" | jq -r '.number')
-            BASE_REF=$(echo "$PR_JSON" | jq -r '.baseRefName')
-            BASE_SHA=$(echo "$PR_JSON" | jq -r '.baseRefOid')
-            HEAD_SHA=$(echo "$PR_JSON" | jq -r '.headRefOid')
-            PR_TITLE=$(echo "$PR_JSON" | jq -r '.title')
-            PR_BODY=$(echo "$PR_JSON" | jq -r '.body // ""')
-            IS_FORK=$(echo "$PR_JSON" | jq -r '.isCrossRepository')
-            PR_AUTHOR=$(echo "$PR_JSON" | jq -r '.author.login // ""')
-          else
-            PR_NUMBER="$EVENT_PR_NUMBER"
-            BASE_REF="$EVENT_BASE_REF"
-            BASE_SHA="$EVENT_BASE_SHA"
-            HEAD_SHA="$EVENT_HEAD_SHA"
-            PR_TITLE="$EVENT_TITLE"
-            PR_BODY="$EVENT_BODY"
-            IS_FORK="$EVENT_FORK"
-            PR_AUTHOR="$EVENT_AUTHOR"
-          fi
-          # Fork PRs run untrusted code with secrets present, so the automatic
-          # pull_request trigger never reviews them. A non-empty INPUT_PR_NUMBER
-          # means we arrived via workflow_call (a maintainer /codex comment gated
-          # by check-write-access), so allow forks only on that path.
-          if [ "$IS_FORK" = "true" ] && [ -z "$INPUT_PR_NUMBER" ]; then
-            echo "Skipping Codex review for fork PR (automatic trigger)."
-            echo "skip=true" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-          # An agent-driven PR flips to ready only after a clean /review round on
-          # a draft, marked by an author comment naming the head SHA (pr skill,
-          # "Review rounds"). Re-reviewing that same head on ready_for_review is
-          # redundant. The marker alone is author attestation, so also require
-          # reviewer evidence: a Codex review (posted by github-actions[bot], not
-          # forgeable by the author) that predates the marker and carries a
-          # non-blocking verdict. Comment-triggered and synchronize runs never
-          # skip. Keep the three copies of this check in sync (pr-ready-review /
-          # codex-pr-review / pi-pr-review); a shared local action would need the
-          # repo checked out before the check, which the fork paths here
-          # deliberately avoid.
-          if [ "$EVENT_ACTION" = "ready_for_review" ] && [ -z "$INPUT_PR_NUMBER" ]; then
-            # Fetch failures fail open (no skip): an API hiccup must run the
-            # review, never skip it or fail the job.
-            COMMENTS=$(gh api "repos/${{ github.repository }}/issues/$PR_NUMBER/comments?per_page=100" --paginate | jq -s '[.[][]]') || COMMENTS='[]'
-            MARKER_TIME=$(jq -r --arg author "$PR_AUTHOR" --arg marker "✅ Review round clean @ $HEAD_SHA" \
-              '[.[] | select(.user.login == $author) | select(.body | contains($marker)) | .created_at] | min // empty' <<<"$COMMENTS")
-            CODEX_VERDICT=''
-            if [ -n "$MARKER_TIME" ]; then
-              # Only Codex evidence that predates the marker counts: the ready-
-              # triggered Codex run itself posts after the flip and must not
-              # vouch for a sibling reviewer's skip.
-              CODEX_VERDICT=$(jq -r --arg mt "$MARKER_TIME" \
-                '[.[] | select(.user.login == "github-actions[bot]") | select(.body | contains("## Codex Review")) | select(.created_at < $mt)] | last | .body // ""' <<<"$COMMENTS" \
-                | grep -m1 -oE '(Good to merge|Mergeable, but should ideally address nits|Should address issues before merging)' || true)
-            fi
-            if [ -n "$MARKER_TIME" ] && [ -n "$CODEX_VERDICT" ] && [ "$CODEX_VERDICT" != "Should address issues before merging" ]; then
-              echo "Clean review round marker found for $HEAD_SHA with pre-marker non-blocking Codex verdict; skipping redundant review."
-              echo "skip=true" >> "$GITHUB_OUTPUT"
-              exit 0
-            fi
-          fi
-          # PR title/body are attacker-controlled free text. Use an unguessable
-          # per-run delimiter so a fork can't embed a fixed heredoc terminator to
-          # inject extra outputs — e.g. is_fork=false (last-write-wins), which
-          # would re-enable the EE checkout and trusted-path settings for forks.
-          RAND=$(head -c 16 /dev/urandom | od -An -tx1 | tr -d ' \n')
-          TITLE_EOF="TITLE_EOF_${RAND}"
-          BODY_EOF="BODY_EOF_${RAND}"
-          {
-            echo "skip=false"
-            echo "is_fork=$IS_FORK"
-            echo "pr_number=$PR_NUMBER"
-            echo "base_ref=$BASE_REF"
-            echo "base_sha=$BASE_SHA"
-            echo "head_sha=$HEAD_SHA"
-            echo "pr_author=$PR_AUTHOR"
-            echo "title<<$TITLE_EOF"
-            printf '%s\n' "$PR_TITLE"
-            echo "$TITLE_EOF"
-            echo "body<<$BODY_EOF"
-            printf '%s\n' "$PR_BODY"
-            echo "$BODY_EOF"
-          } >> "$GITHUB_OUTPUT"
-
       - name: Checkout repository
-        if: steps.codex_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
+        if: steps.codex_config.outputs.enabled == 'true'
         uses: actions/checkout@v5
         with:
-          ref: refs/pull/${{ steps.pr.outputs.pr_number }}/merge
+          ref: refs/pull/${{ github.event.pull_request.number }}/merge
           fetch-depth: 1
-          # Don't persist github.token in .git/config: the review agent can read
-          # the checkout, and on the fork path that token (issue/PR write) would
-          # otherwise be exfiltratable. All later git ops target the public origin
-          # and need no auth; EE checkout and gh use their own explicit tokens.
-          persist-credentials: false
-
-      # Never expose the EE private-repo token to untrusted fork code. Skipping
-      # this step leaves steps.ee.outputs.available empty, so the EE checkout and
-      # substitution steps below are skipped too.
-      - name: Check EE access
-        if: steps.codex_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true' && steps.pr.outputs.is_fork != 'true'
-        id: ee
-        env:
-          EE_TOKEN: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-        run: |
-          if [ -n "$EE_TOKEN" ]; then
-            echo "available=true" >> "$GITHUB_OUTPUT"
-            echo "ee_repo_ref=$(cat ./backend/ee-repo-ref.txt)" >> "$GITHUB_OUTPUT"
-          else
-            echo "available=false" >> "$GITHUB_OUTPUT"
-          fi
-
-      - name: Checkout EE repository
-        if: steps.ee.outputs.available == 'true'
-        uses: actions/checkout@v5
-        with:
-          repository: windmill-labs/windmill-ee-private
-          path: ./windmill-ee-private
-          ref: ${{ steps.ee.outputs.ee_repo_ref }}
-          token: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-          fetch-depth: 1
-
-      - name: Substitute EE code
-        if: steps.ee.outputs.available == 'true'
-        run: ./backend/substitute_ee_code.sh --copy --dir ./windmill-ee-private
 
       - name: Set up Node.js
-        if: steps.codex_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
+        if: steps.codex_config.outputs.enabled == 'true'
         uses: actions/setup-node@v4
         with:
           node-version: 22
 
       - name: Install Codex CLI
-        if: steps.codex_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        run: npm install --global @openai/codex@0.144.1
+        if: steps.codex_config.outputs.enabled == 'true'
+        run: npm install --global @openai/codex@0.117.0
 
-      - name: Configure Codex auth
-        if: steps.codex_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
+      - name: Configure file-backed Codex auth
+        if: steps.codex_config.outputs.enabled == 'true'
         env:
-          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
           CODEX_AUTH_JSON: ${{ secrets.CODEX_AUTH_JSON }}
         run: |
           CODEX_HOME="$HOME/.codex"
@@ -234,61 +58,36 @@ jobs:
           cat > "$CODEX_HOME/config.toml" <<'EOF'
           cli_auth_credentials_store = "file"
           EOF
-          if [ -n "$OPENAI_API_KEY" ]; then
-            printf '%s' "$OPENAI_API_KEY" | codex login --with-api-key
-          else
-            printf '%s' "$CODEX_AUTH_JSON" > "$CODEX_HOME/auth.json"
-            chmod 600 "$CODEX_HOME/auth.json"
-            node -e 'JSON.parse(require("fs").readFileSync(process.argv[1], "utf8"))' "$CODEX_HOME/auth.json"
-          fi
+          printf '%s' "$CODEX_AUTH_JSON" > "$CODEX_HOME/auth.json"
+          chmod 600 "$CODEX_HOME/auth.json"
+          node -e 'JSON.parse(require("fs").readFileSync(process.argv[1], "utf8"))' "$CODEX_HOME/auth.json"
 
       - name: Pre-fetch base and head refs for the PR
-        if: steps.codex_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
+        if: steps.codex_config.outputs.enabled == 'true'
         env:
-          PR_BASE_REF: ${{ steps.pr.outputs.base_ref }}
-          PR_NUMBER: ${{ steps.pr.outputs.pr_number }}
+          PR_BASE_REF: ${{ github.event.pull_request.base.ref }}
+          PR_NUMBER: ${{ github.event.pull_request.number }}
         run: |
           git fetch --no-tags origin \
             "$PR_BASE_REF" \
             "+refs/pull/$PR_NUMBER/head"
 
-      - name: Fetch prior PR discussion
-        if: steps.codex_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        env:
-          GH_TOKEN: ${{ github.token }}
-          REPO: ${{ github.repository }}
-          PR_NUMBER: ${{ steps.pr.outputs.pr_number }}
-        run: |
-          # Write outside the checkout: on the fork path the merge tree is
-          # attacker-controlled, and a committed symlink at this path would
-          # redirect the write.
-          gh api "repos/$REPO/issues/$PR_NUMBER/comments?per_page=100" \
-            --jq '[.[] | {user: .user.login, created_at: .created_at, body: (.body | .[:4000])}] | sort_by(.created_at) | .[-20:]' \
-            > "$RUNNER_TEMP/prior-comments.json" || echo "[]" > "$RUNNER_TEMP/prior-comments.json"
-
       - name: Write Codex review context
-        if: steps.codex_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
+        if: steps.codex_config.outputs.enabled == 'true'
         env:
           PR_REPOSITORY: ${{ github.repository }}
-          PR_NUMBER: ${{ steps.pr.outputs.pr_number }}
-          PR_BASE_SHA: ${{ steps.pr.outputs.base_sha }}
-          PR_HEAD_SHA: ${{ steps.pr.outputs.head_sha }}
-          PR_TITLE: ${{ steps.pr.outputs.title }}
-          PR_BODY: ${{ steps.pr.outputs.body }}
-          PR_AUTHOR: ${{ steps.pr.outputs.pr_author }}
-          EXTRA_PROMPT: ${{ inputs.extra_prompt }}
+          PR_NUMBER: ${{ github.event.pull_request.number }}
+          PR_BASE_SHA: ${{ github.event.pull_request.base.sha }}
+          PR_HEAD_SHA: ${{ github.event.pull_request.head.sha }}
+          PR_TITLE: ${{ github.event.pull_request.title }}
+          PR_BODY: ${{ github.event.pull_request.body || '' }}
         run: |
+          mkdir -p .github/codex
           node <<'NODE'
           const fs = require('fs');
-          const tmp = process.env.RUNNER_TEMP;
           const lines = [
             `Repository: ${process.env.PR_REPOSITORY}`,
             `PR number: ${process.env.PR_NUMBER}`,
-          ];
-          if (process.env.PR_AUTHOR) {
-            lines.push(`PR AUTHOR: ${process.env.PR_AUTHOR}`);
-          }
-          lines.push(
             `Base SHA: ${process.env.PR_BASE_SHA}`,
             `Head SHA: ${process.env.PR_HEAD_SHA}`,
             '',
@@ -306,116 +105,41 @@ jobs:
             '',
             'Full review diff command:',
             `git diff --unified=0 ${process.env.PR_BASE_SHA}...${process.env.PR_HEAD_SHA}`
-          );
-          if (process.env.EXTRA_PROMPT && process.env.EXTRA_PROMPT.trim()) {
-            lines.push('', 'Additional reviewer instructions:', process.env.EXTRA_PROMPT.trim());
-          }
-          if (fs.existsSync(`${tmp}/prior-comments.json`)) {
-            try {
-              const comments = JSON.parse(fs.readFileSync(`${tmp}/prior-comments.json`, 'utf8'));
-              if (Array.isArray(comments) && comments.length > 0) {
-                lines.push(
-                  '',
-                  'Prior PR discussion (most recent up to 20 comments):',
-                  '',
-                  'If you have already reviewed this PR (look for your own earlier "## Codex Review" comment), focus on what changed since then per the diff and respect any decisions the human made in replies. Do not re-flag findings the human already pushed back on.',
-                  ''
-                );
-                for (const c of comments) {
-                  lines.push(`### @${c.user} (${c.created_at})`, '', c.body, '', '---', '');
-                }
-              }
-            } catch (_) {}
-          }
-          fs.writeFileSync(`${tmp}/pr-review-context.md`, `${lines.join('\n')}\n`);
+          ];
+          fs.writeFileSync('.github/codex/pr-review-context.md', `${lines.join('\n')}\n`);
           NODE
 
       - name: Run Codex review
-        if: steps.codex_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        env:
-          PR_IS_FORK: ${{ steps.pr.outputs.is_fork }}
-          PR_BASE_REF: ${{ steps.pr.outputs.base_ref }}
+        if: steps.codex_config.outputs.enabled == 'true'
         run: |
-          if [ "$PR_IS_FORK" = "true" ]; then
-            # Fork code is untrusted. Read the review policy/prompt from the base
-            # ref (git show) rather than the attacker-controlled merge checkout,
-            # so a malicious fork can't rewrite the reviewer's own instructions,
-            # and run in a network-disabled sandbox to block secret exfiltration.
-            git show "origin/$PR_BASE_REF:REVIEW.md" > /tmp/codex-prompt.md
-            git show "origin/$PR_BASE_REF:.github/codex/pr-review.prompt.md" >> /tmp/codex-prompt.md
-            SANDBOX_MODE=workspace-write
-          else
-            cat REVIEW.md .github/codex/pr-review.prompt.md > /tmp/codex-prompt.md
-            SANDBOX_MODE=danger-full-access
-          fi
-          # The context file lives in RUNNER_TEMP (outside the attacker-controlled
-          # checkout); tell the agent its absolute path.
-          printf '\nReview context file (absolute path): %s\n' "$RUNNER_TEMP/pr-review-context.md" >> /tmp/codex-prompt.md
-          # Write the final message outside the checkout too: a fork could commit
-          # codex-final-message.md as a symlink and redirect this write to overwrite
-          # e.g. a GitHub Action's index.js, which then runs with our credentials.
           codex exec \
             -C "$GITHUB_WORKSPACE" \
-            -m gpt-5.6-sol \
+            -m gpt-5.4 \
             -c 'model_reasoning_effort="xhigh"' \
-            -s "$SANDBOX_MODE" \
-            -o "$RUNNER_TEMP/codex-final-message.md" \
-            - < /tmp/codex-prompt.md
+            -s read-only \
+            -o codex-final-message.md \
+            - < .github/codex/pr-review.prompt.md
 
       - name: Post Codex review comment
-        if: steps.codex_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
+        if: steps.codex_config.outputs.enabled == 'true'
         uses: actions/github-script@v7
-        env:
-          PR_NUMBER: ${{ steps.pr.outputs.pr_number }}
-          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-          CODEX_AUTH_JSON: ${{ secrets.CODEX_AUTH_JSON }}
-          GH_JOB_TOKEN: ${{ github.token }}
         with:
           github-token: ${{ github.token }}
           script: |
             const fs = require('fs');
-            const path = `${process.env.RUNNER_TEMP}/codex-final-message.md`;
+            const path = `${process.env.GITHUB_WORKSPACE}/codex-final-message.md`;
             if (!fs.existsSync(path)) {
               core.info('Codex did not produce a final message; skipping PR comment.');
               return;
             }
-            let body = fs.readFileSync(path, 'utf8').trim();
+            const body = fs.readFileSync(path, 'utf8').trim();
             if (!body) {
               core.info('Codex final message was empty; skipping PR comment.');
               return;
             }
-            // Defense-in-depth for fork reviews: the model call needs the provider
-            // credential in the env, and the posted comment bypasses Actions log
-            // masking. Strip any credential (API key, raw auth JSON, nested
-            // tokens) that leaked into the review text before posting.
-            const secrets = [];
-            const addSecret = (v, min) => {
-              if (typeof v === 'string' && v.length >= min) secrets.push(v);
-            };
-            addSecret(process.env.OPENAI_API_KEY, 8);
-            addSecret(process.env.CODEX_AUTH_JSON, 8);
-            addSecret(process.env.GH_JOB_TOKEN, 8);
-            if (process.env.CODEX_AUTH_JSON) {
-              try {
-                const collect = (o) => {
-                  if (typeof o === 'string') addSecret(o, 20);
-                  else if (Array.isArray(o)) o.forEach(collect);
-                  else if (o && typeof o === 'object') Object.values(o).forEach(collect);
-                };
-                collect(JSON.parse(process.env.CODEX_AUTH_JSON));
-              } catch (_) {}
-            }
-            for (const s of [...new Set(secrets)].sort((a, b) => b.length - a.length)) {
-              body = body.split(s).join('[REDACTED]');
-            }
-            body = body.trim();
-            if (!body) {
-              core.info('Codex final message was empty after redaction; skipping PR comment.');
-              return;
-            }
             await github.rest.issues.createComment({
               owner: context.repo.owner,
               repo: context.repo.repo,
-              issue_number: Number(process.env.PR_NUMBER),
+              issue_number: context.payload.pull_request.number,
               body,
             });

.github/workflows/docker-image-rpi4.yml

@@ -68,7 +68,6 @@ jobs:
           push: true
           build-args: |
             features=ce_rpi
-            WM_BUILD_VERSION=${{ github.sha }}
           tags: |
             ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:dev
             ${{ steps.meta-public.outputs.tags }}

.github/workflows/docker-image.yml

@@ -93,7 +93,6 @@ jobs:
           push: true
           build-args: |
             features=ce
-            WM_BUILD_VERSION=${{ github.sha }}
           tags: |
             ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}:${{ env.DEV_SHA }}
             ${{ steps.meta-public.outputs.tags }}
@@ -156,7 +155,6 @@ jobs:
           push: true
           build-args: |
             features=ee
-            WM_BUILD_VERSION=${{ github.sha }}
           tags: |
             ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}-ee:${{ env.DEV_SHA }}
             ${{ steps.meta-ee-public.outputs.tags }}
@@ -256,7 +254,6 @@ jobs:
           target: debuginfo
           build-args: |
             features=ee
-            WM_BUILD_VERSION=${{ github.sha }}
           outputs: type=local,dest=./debuginfo
 
       - name: Rename debug file with corresponding architecture

.github/workflows/frontend-check.yml

@@ -23,8 +23,5 @@ jobs:
           cache-dependency-path: "frontend/package-lock.json"
       - name: "npm check"
         timeout-minutes: 5
-        env:
-          # svelte-check peaks past node's ~4GB default ceiling on this runner and aborts.
-          NODE_OPTIONS: --max-old-space-size=8192
         run: cd frontend && npm ci && npm run generate-backend-client && npm run
           check

.github/workflows/git-commands.yaml

@@ -5,22 +5,21 @@ on:
     types: [created]
 
 jobs:
-  # /command comments can come from anyone; author_association misses private org
-  # members, so check-access resolves them via the internal app token. Runs once and is
-  # OR'd into each job's guard (public members still pass on author_association alone).
-  check-access:
-    if: github.event.issue.pull_request != null && startsWith(github.event.comment.body, '/')
-    uses: ./.github/workflows/check-write-access.yml
-    with:
-      username: ${{ github.event.comment.user.login }}
-    secrets: inherit
+  check-membership:
+    if: >-
+      github.event.issue.pull_request && (
+        startsWith(github.event.comment.body, '/updatesqlx') ||
+        startsWith(github.event.comment.body, '/demo') ||
+        startsWith(github.event.comment.body, '/eeref') ||
+        startsWith(github.event.comment.body, '/docs')
+      )
+    uses: ./.github/workflows/check-org-membership.yml
+    secrets:
+      access_token: ${{ secrets.ORG_ACCESS_TOKEN }}
 
   update-sqlx:
-    needs: [check-access]
-    if: >-
-      github.event.issue.pull_request &&
-      (contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association) || needs.check-access.outputs.authorized == 'true') &&
-      startsWith(github.event.comment.body, '/updatesqlx')
+    needs: check-membership
+    if: needs.check-membership.outputs.is_member == 'true' && startsWith(github.event.comment.body, '/updatesqlx')
     runs-on: ubicloud-standard-8
     permissions:
       contents: write
@@ -80,7 +79,7 @@ jobs:
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           cache-workspaces: backend
-          toolchain: 1.97.0
+          toolchain: 1.93.0
 
       - name: Install xmlsec and gssapi build-time deps
         run: |
@@ -148,11 +147,8 @@ jobs:
             })
 
   demo:
-    needs: [check-access]
-    if: >-
-      github.event.issue.pull_request &&
-      (contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association) || needs.check-access.outputs.authorized == 'true') &&
-      startsWith(github.event.comment.body, '/demo')
+    needs: check-membership
+    if: needs.check-membership.outputs.is_member == 'true' && startsWith(github.event.comment.body, '/demo')
     runs-on: ubicloud-standard-2
     permissions:
       contents: read
@@ -231,11 +227,8 @@ jobs:
           fi
 
   update-ee-ref:
-    needs: [check-access]
-    if: >-
-      github.event.issue.pull_request &&
-      (contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association) || needs.check-access.outputs.authorized == 'true') &&
-      startsWith(github.event.comment.body, '/eeref')
+    needs: check-membership
+    if: needs.check-membership.outputs.is_member == 'true' && startsWith(github.event.comment.body, '/eeref')
     runs-on: ubicloud-standard-2
     permissions:
       contents: write
@@ -320,11 +313,8 @@ jobs:
             })
 
   update-docs:
-    needs: [check-access]
-    if: >-
-      github.event.issue.pull_request &&
-      (contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association) || needs.check-access.outputs.authorized == 'true') &&
-      startsWith(github.event.comment.body, '/docs')
+    needs: check-membership
+    if: needs.check-membership.outputs.is_member == 'true' && startsWith(github.event.comment.body, '/docs')
     runs-on: ubicloud-standard-2
     permissions:
       contents: read

.github/workflows/git-sync-test.yml

@@ -8,11 +8,6 @@ on:
       - "backend/windmill-git-sync/**"
       - "backend/windmill-api-integration-tests/tests/git_sync*"
       - "backend/ee-repo-ref.txt"
-      - "backend/windmill-common/src/workspaces.rs"
-      - "backend/windmill-worker/src/result_processor.rs"
-      - "backend/windmill-api-workspaces/**"
-      - "cli/src/commands/sync/**"
-      - "cli/src/utils/git.ts"
       - "integration_tests/test/git_sync_test.py"
       - ".github/workflows/git-sync-test.yml"
   pull_request:
@@ -21,11 +16,6 @@ on:
       - "backend/windmill-git-sync/**"
       - "backend/windmill-api-integration-tests/tests/git_sync*"
       - "backend/ee-repo-ref.txt"
-      - "backend/windmill-common/src/workspaces.rs"
-      - "backend/windmill-worker/src/result_processor.rs"
-      - "backend/windmill-api-workspaces/**"
-      - "cli/src/commands/sync/**"
-      - "cli/src/utils/git.ts"
       - "integration_tests/test/git_sync_test.py"
       - ".github/workflows/git-sync-test.yml"
 
@@ -58,8 +48,8 @@ jobs:
           echo "Changed files:"
           echo "$CHANGED_FILES"
 
-          # Direct git sync file changes — always relevant.
-          if echo "$CHANGED_FILES" | grep -qE '^(backend/windmill-git-sync/|backend/windmill-worker/src/result_processor\.rs|backend/windmill-api-workspaces/|backend/windmill-api-integration-tests/tests/git_sync|backend/windmill-common/src/workspaces\.rs|cli/src/commands/sync/|cli/src/utils/git\.ts|integration_tests/test/git_sync|\.github/workflows/git-sync-test\.yml)'; then
+          # Direct git sync file changes — always relevant
+          if echo "$CHANGED_FILES" | grep -qE '^(backend/windmill-git-sync/|backend/windmill-api-integration-tests/tests/git_sync|integration_tests/test/git_sync|\.github/workflows/git-sync-test\.yml)'; then
             echo "should_run=true" >> "$GITHUB_OUTPUT"
             echo "Relevant: direct git sync file changes"
             exit 0
@@ -129,7 +119,7 @@ jobs:
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           cache-workspaces: backend
-          toolchain: 1.97.0
+          toolchain: 1.93.0
 
       - uses: oven-sh/setup-bun@v2
         with:
@@ -188,9 +178,6 @@ jobs:
           DENO_PATH: deno
           BUN_PATH: bun
           NODE_BIN_PATH: node
-          # The auto-pull poller's SSRF guard rejects localhost git remotes;
-          # the tests' Gitea runs on localhost.
-          ALLOW_LOCAL_GIT_REMOTES: "true"
         run: |
           ./target/debug/windmill &
           echo "Waiting for Windmill to be ready..."

.github/workflows/go_on_release.yml

@@ -27,7 +27,7 @@ jobs:
           go build
       - name: Pushes to another repository
         id: push_directory
-        uses: cpina/github-action-push-to-another-repository@55306faa4ed53b815ae49e564af8cfb359d32ae2 # v1.7.3
+        uses: cpina/github-action-push-to-another-repository@devel
         env:
           API_TOKEN_GITHUB: ${{ secrets.DENO_PAT }}
         with:

.github/workflows/pi-pr-review.yml

@@ -1,442 +0,0 @@
-name: Pi Auto Review
-
-on:
-  pull_request:
-    types: [ready_for_review, opened, synchronize]
-  workflow_call:
-    inputs:
-      pr_number:
-        description: 'PR number to review'
-        required: true
-        type: number
-      extra_prompt:
-        description: 'Additional reviewer instructions appended to the standard review prompt'
-        required: false
-        type: string
-        default: ''
-      triggered_by:
-        description: 'GitHub username that triggered this review (for audit only)'
-        required: false
-        type: string
-        default: ''
-    secrets:
-      DEEPSEEK_API_KEY:
-        required: false
-      WINDMILL_EE_PRIVATE_ACCESS:
-        required: false
-
-concurrency:
-  group: pi-review-${{ inputs.pr_number || github.event.pull_request.number }}
-  cancel-in-progress: true
-
-jobs:
-  pi-review:
-    runs-on: ubicloud-standard-2
-    timeout-minutes: 30
-    # A non-fork PR (head.repo.fork == false) can only be opened by someone with push
-    # access to this repo, so fork==false already enforces write access. Do NOT re-add
-    # an author_association gate: the pull_request webhook payload reports private org
-    # members as CONTRIBUTOR/NONE (only public members show as MEMBER), which silently
-    # skips auto-review for every private member.
-    if: |
-      github.event_name == 'workflow_call' ||
-      (
-        github.event.pull_request.draft == false &&
-        github.event.pull_request.head.repo.fork == false
-      )
-    permissions:
-      contents: read
-      issues: write
-      pull-requests: write
-    steps:
-      - name: Check Pi configuration
-        id: pi_config
-        env:
-          DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}
-        run: |
-          if [ -n "$DEEPSEEK_API_KEY" ]; then
-            echo "enabled=true" >> "$GITHUB_OUTPUT"
-          else
-            echo "enabled=false" >> "$GITHUB_OUTPUT"
-            echo "DEEPSEEK_API_KEY is not configured; skipping Pi review."
-          fi
-
-      - name: Resolve PR metadata
-        if: steps.pi_config.outputs.enabled == 'true'
-        id: pr
-        env:
-          GH_TOKEN: ${{ github.token }}
-          INPUT_PR_NUMBER: ${{ inputs.pr_number }}
-          EVENT_PR_NUMBER: ${{ github.event.pull_request.number }}
-          EVENT_BASE_REF: ${{ github.event.pull_request.base.ref }}
-          EVENT_BASE_SHA: ${{ github.event.pull_request.base.sha }}
-          EVENT_HEAD_SHA: ${{ github.event.pull_request.head.sha }}
-          EVENT_TITLE: ${{ github.event.pull_request.title }}
-          EVENT_BODY: ${{ github.event.pull_request.body }}
-          EVENT_FORK: ${{ github.event.pull_request.head.repo.fork }}
-          EVENT_AUTHOR: ${{ github.event.pull_request.user.login }}
-          EVENT_ACTION: ${{ github.event.action }}
-        run: |
-          if [ -n "$INPUT_PR_NUMBER" ]; then
-            PR_JSON=$(gh pr view "$INPUT_PR_NUMBER" --repo "${{ github.repository }}" \
-              --json number,baseRefName,baseRefOid,headRefOid,title,body,isCrossRepository,author)
-            PR_NUMBER=$(echo "$PR_JSON" | jq -r '.number')
-            BASE_REF=$(echo "$PR_JSON" | jq -r '.baseRefName')
-            BASE_SHA=$(echo "$PR_JSON" | jq -r '.baseRefOid')
-            HEAD_SHA=$(echo "$PR_JSON" | jq -r '.headRefOid')
-            PR_TITLE=$(echo "$PR_JSON" | jq -r '.title')
-            PR_BODY=$(echo "$PR_JSON" | jq -r '.body // ""')
-            IS_FORK=$(echo "$PR_JSON" | jq -r '.isCrossRepository')
-            PR_AUTHOR=$(echo "$PR_JSON" | jq -r '.author.login // ""')
-          else
-            PR_NUMBER="$EVENT_PR_NUMBER"
-            BASE_REF="$EVENT_BASE_REF"
-            BASE_SHA="$EVENT_BASE_SHA"
-            HEAD_SHA="$EVENT_HEAD_SHA"
-            PR_TITLE="$EVENT_TITLE"
-            PR_BODY="$EVENT_BODY"
-            IS_FORK="$EVENT_FORK"
-            PR_AUTHOR="$EVENT_AUTHOR"
-          fi
-          # Fork PRs run untrusted code with secrets present, so the automatic
-          # pull_request trigger never reviews them. A non-empty INPUT_PR_NUMBER
-          # means we arrived via workflow_call (a maintainer /pi comment gated by
-          # check-write-access), so allow forks only on that path.
-          if [ "$IS_FORK" = "true" ] && [ -z "$INPUT_PR_NUMBER" ]; then
-            echo "Skipping Pi review for fork PR (automatic trigger)."
-            echo "skip=true" >> "$GITHUB_OUTPUT"
-            exit 0
-          fi
-          # An agent-driven PR flips to ready only after a clean /review round on
-          # a draft, marked by an author comment naming the head SHA (pr skill,
-          # "Review rounds"). Re-reviewing that same head on ready_for_review is
-          # redundant. The marker alone is author attestation, so also require
-          # reviewer evidence: a Codex review (posted by github-actions[bot], not
-          # forgeable by the author) that predates the marker and carries a
-          # non-blocking verdict. Comment-triggered and synchronize runs never
-          # skip. Keep the three copies of this check in sync (pr-ready-review /
-          # codex-pr-review / pi-pr-review); a shared local action would need the
-          # repo checked out before the check, which the fork paths here
-          # deliberately avoid.
-          if [ "$EVENT_ACTION" = "ready_for_review" ] && [ -z "$INPUT_PR_NUMBER" ]; then
-            # Fetch failures fail open (no skip): an API hiccup must run the
-            # review, never skip it or fail the job.
-            COMMENTS=$(gh api "repos/${{ github.repository }}/issues/$PR_NUMBER/comments?per_page=100" --paginate | jq -s '[.[][]]') || COMMENTS='[]'
-            MARKER_TIME=$(jq -r --arg author "$PR_AUTHOR" --arg marker "✅ Review round clean @ $HEAD_SHA" \
-              '[.[] | select(.user.login == $author) | select(.body | contains($marker)) | .created_at] | min // empty' <<<"$COMMENTS")
-            CODEX_VERDICT=''
-            if [ -n "$MARKER_TIME" ]; then
-              # Only Codex evidence that predates the marker counts: the ready-
-              # triggered Codex run itself posts after the flip and must not
-              # vouch for a sibling reviewer's skip.
-              CODEX_VERDICT=$(jq -r --arg mt "$MARKER_TIME" \
-                '[.[] | select(.user.login == "github-actions[bot]") | select(.body | contains("## Codex Review")) | select(.created_at < $mt)] | last | .body // ""' <<<"$COMMENTS" \
-                | grep -m1 -oE '(Good to merge|Mergeable, but should ideally address nits|Should address issues before merging)' || true)
-            fi
-            if [ -n "$MARKER_TIME" ] && [ -n "$CODEX_VERDICT" ] && [ "$CODEX_VERDICT" != "Should address issues before merging" ]; then
-              echo "Clean review round marker found for $HEAD_SHA with pre-marker non-blocking Codex verdict; skipping redundant review."
-              echo "skip=true" >> "$GITHUB_OUTPUT"
-              exit 0
-            fi
-          fi
-          # PR title/body are attacker-controlled free text. Use an unguessable
-          # per-run delimiter so a fork can't embed a fixed heredoc terminator to
-          # inject extra outputs — e.g. is_fork=false (last-write-wins), which
-          # would re-enable the EE checkout and trusted-path settings for forks.
-          RAND=$(head -c 16 /dev/urandom | od -An -tx1 | tr -d ' \n')
-          TITLE_EOF="TITLE_EOF_${RAND}"
-          BODY_EOF="BODY_EOF_${RAND}"
-          {
-            echo "skip=false"
-            echo "is_fork=$IS_FORK"
-            echo "pr_number=$PR_NUMBER"
-            echo "base_ref=$BASE_REF"
-            echo "base_sha=$BASE_SHA"
-            echo "head_sha=$HEAD_SHA"
-            echo "pr_author=$PR_AUTHOR"
-            echo "title<<$TITLE_EOF"
-            printf '%s\n' "$PR_TITLE"
-            echo "$TITLE_EOF"
-            echo "body<<$BODY_EOF"
-            printf '%s\n' "$PR_BODY"
-            echo "$BODY_EOF"
-          } >> "$GITHUB_OUTPUT"
-
-      - name: Checkout repository
-        if: steps.pi_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        uses: actions/checkout@v5
-        with:
-          ref: refs/pull/${{ steps.pr.outputs.pr_number }}/merge
-          fetch-depth: 1
-          # Don't persist github.token in .git/config: the review agent can read
-          # the checkout, and on the fork path that token (issue/PR write) would
-          # otherwise be exfiltratable. All later git ops target the public origin
-          # and need no auth; EE checkout and gh use their own explicit tokens.
-          persist-credentials: false
-
-      # Never expose the EE private-repo token to untrusted fork code. Skipping
-      # this step leaves steps.ee.outputs.available empty, so the EE checkout and
-      # substitution steps below are skipped too.
-      - name: Check EE access
-        if: steps.pi_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true' && steps.pr.outputs.is_fork != 'true'
-        id: ee
-        env:
-          EE_TOKEN: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-        run: |
-          if [ -n "$EE_TOKEN" ]; then
-            echo "available=true" >> "$GITHUB_OUTPUT"
-            echo "ee_repo_ref=$(cat ./backend/ee-repo-ref.txt)" >> "$GITHUB_OUTPUT"
-          else
-            echo "available=false" >> "$GITHUB_OUTPUT"
-          fi
-
-      - name: Checkout EE repository
-        if: steps.ee.outputs.available == 'true'
-        uses: actions/checkout@v5
-        with:
-          repository: windmill-labs/windmill-ee-private
-          path: ./windmill-ee-private
-          ref: ${{ steps.ee.outputs.ee_repo_ref }}
-          token: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-          fetch-depth: 1
-
-      - name: Substitute EE code
-        if: steps.ee.outputs.available == 'true'
-        run: ./backend/substitute_ee_code.sh --copy --dir ./windmill-ee-private
-
-      - name: Set up Node.js
-        if: steps.pi_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        uses: actions/setup-node@v4
-        with:
-          node-version: 22
-
-      - name: Install Pi CLI
-        if: steps.pi_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        run: npm install --global @mariozechner/pi-coding-agent
-
-      - name: Pre-fetch base and head refs for the PR
-        if: steps.pi_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        env:
-          PR_BASE_REF: ${{ steps.pr.outputs.base_ref }}
-          PR_NUMBER: ${{ steps.pr.outputs.pr_number }}
-        run: |
-          git fetch --no-tags origin \
-            "$PR_BASE_REF" \
-            "+refs/pull/$PR_NUMBER/head"
-
-      - name: Fetch prior PR discussion
-        if: steps.pi_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        env:
-          GH_TOKEN: ${{ github.token }}
-          REPO: ${{ github.repository }}
-          PR_NUMBER: ${{ steps.pr.outputs.pr_number }}
-        run: |
-          # Write outside the checkout: on the fork path the merge tree is
-          # attacker-controlled, and a committed symlink at this path would
-          # redirect the write.
-          gh api "repos/$REPO/issues/$PR_NUMBER/comments?per_page=100" \
-            --jq '[.[] | {user: .user.login, created_at: .created_at, body: (.body | .[:4000])}] | sort_by(.created_at) | .[-20:]' \
-            > "$RUNNER_TEMP/prior-comments.json" || echo "[]" > "$RUNNER_TEMP/prior-comments.json"
-
-      - name: Write Pi review context
-        if: steps.pi_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        env:
-          PR_REPOSITORY: ${{ github.repository }}
-          PR_NUMBER: ${{ steps.pr.outputs.pr_number }}
-          PR_BASE_SHA: ${{ steps.pr.outputs.base_sha }}
-          PR_HEAD_SHA: ${{ steps.pr.outputs.head_sha }}
-          PR_TITLE: ${{ steps.pr.outputs.title }}
-          PR_BODY: ${{ steps.pr.outputs.body }}
-          PR_AUTHOR: ${{ steps.pr.outputs.pr_author }}
-          EXTRA_PROMPT: ${{ inputs.extra_prompt }}
-        run: |
-          node <<'NODE'
-          const fs = require('fs');
-          const tmp = process.env.RUNNER_TEMP;
-          const lines = [
-            `Repository: ${process.env.PR_REPOSITORY}`,
-            `PR number: ${process.env.PR_NUMBER}`,
-          ];
-          if (process.env.PR_AUTHOR) {
-            lines.push(`PR AUTHOR: ${process.env.PR_AUTHOR}`);
-          }
-          lines.push(
-            `Base SHA: ${process.env.PR_BASE_SHA}`,
-            `Head SHA: ${process.env.PR_HEAD_SHA}`,
-            '',
-            'PR title:',
-            process.env.PR_TITLE || '(empty)',
-            '',
-            'PR body:',
-            process.env.PR_BODY || '(empty)',
-            '',
-            'Changed commits command:',
-            `git log --oneline ${process.env.PR_BASE_SHA}...${process.env.PR_HEAD_SHA}`,
-            '',
-            'Changed files command:',
-            `git diff --stat ${process.env.PR_BASE_SHA}...${process.env.PR_HEAD_SHA}`,
-            '',
-            'Full review diff command:',
-            `git diff --unified=0 ${process.env.PR_BASE_SHA}...${process.env.PR_HEAD_SHA}`
-          );
-          if (process.env.EXTRA_PROMPT && process.env.EXTRA_PROMPT.trim()) {
-            lines.push('', 'Additional reviewer instructions:', process.env.EXTRA_PROMPT.trim());
-          }
-          if (fs.existsSync(`${tmp}/prior-comments.json`)) {
-            try {
-              const comments = JSON.parse(fs.readFileSync(`${tmp}/prior-comments.json`, 'utf8'));
-              if (Array.isArray(comments) && comments.length > 0) {
-                lines.push(
-                  '',
-                  'Prior PR discussion (most recent up to 20 comments):',
-                  '',
-                  'If you have already reviewed this PR (look for your own earlier "## Pi Review (DeepSeek V4)" comment), focus on what changed since then per the diff and respect any decisions the human made in replies. Do not re-flag findings the human already pushed back on.',
-                  ''
-                );
-                for (const c of comments) {
-                  lines.push(`### @${c.user} (${c.created_at})`, '', c.body, '', '---', '');
-                }
-              }
-            } catch (_) {}
-          }
-          fs.writeFileSync(`${tmp}/pr-review-context.md`, `${lines.join('\n')}\n`);
-          NODE
-
-      - name: Run Pi review
-        if: steps.pi_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        env:
-          DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}
-          PI_SKIP_VERSION_CHECK: '1'
-          PR_IS_FORK: ${{ steps.pr.outputs.is_fork }}
-          PR_BASE_REF: ${{ steps.pr.outputs.base_ref }}
-          PR_BASE_SHA: ${{ steps.pr.outputs.base_sha }}
-          PR_HEAD_SHA: ${{ steps.pr.outputs.head_sha }}
-        run: |
-          set -o pipefail
-          PI_HARDEN_FLAGS=()
-          # Keep generated files (final message, events, context) outside the
-          # checkout: on the fork path a committed symlink at any of these paths
-          # would redirect our write and could overwrite an action's code that
-          # then runs with our credentials. RUNNER_TEMP is outside the checkout.
-          OUT_DIR="$RUNNER_TEMP"
-          CTX="$RUNNER_TEMP/pr-review-context.md"
-          if [ "$PR_IS_FORK" = "true" ]; then
-            # Fork code is untrusted. Read the review policy/prompt from the base
-            # ref (git show) rather than the attacker-controlled merge checkout,
-            # so a malicious fork can't rewrite the reviewer's own instructions,
-            # and drop the bash tool so the agent has no shell to exfiltrate with.
-            git show "origin/$PR_BASE_REF:REVIEW.md" > /tmp/pi-prompt.md
-            git show "origin/$PR_BASE_REF:.github/pi/pr-review.prompt.md" >> /tmp/pi-prompt.md
-            PI_TOOLS=read,grep,find,ls
-
-            # The agent has no shell, so pre-compute the diff (base...head SHAs are
-            # trusted) into the context file it reads. It may still read fork files
-            # by absolute path for extra context — reads are safe.
-            {
-              echo ""
-              echo "## Pre-computed review diff (base...head)"
-              echo "You have no shell. The full diff is below. The repository checkout"
-              echo "is at $GITHUB_WORKSPACE — you may read files there by absolute path."
-              echo '```diff'
-              git -C "$GITHUB_WORKSPACE" diff --unified=0 "$PR_BASE_SHA...$PR_HEAD_SHA"
-              echo '```'
-            } >> "$CTX"
-
-            # Pi resolves ALL project config from <cwd>/.pi (settings/packages,
-            # extensions, skills, themes, prompts, SYSTEM.md); inside the fork
-            # checkout a fork could inject any to run code or rewrite our system
-            # prompt. Discovery is cwd-based, so run from a fresh empty dir.
-            PI_WORKDIR=$(mktemp -d)
-            cd "$PI_WORKDIR"
-
-            # Belt-and-suspenders on top of the isolated cwd: refuse discovery of
-            # extensions/skills/templates/themes/context-files, and PI_OFFLINE=1 to
-            # block any startup network op or package install. PI_OFFLINE gates only
-            # startup network ops, not the provider inference call.
-            PI_HARDEN_FLAGS=(--no-extensions --no-skills --no-prompt-templates --no-themes --no-context-files)
-            export PI_OFFLINE=1
-          else
-            cat REVIEW.md .github/pi/pr-review.prompt.md > /tmp/pi-prompt.md
-            PI_TOOLS=read,grep,find,ls,bash
-          fi
-          # The context file lives in RUNNER_TEMP (outside the checkout); tell the
-          # agent its absolute path.
-          printf '\nReview context file (absolute path): %s\n' "$CTX" >> /tmp/pi-prompt.md
-          pi -p \
-            --provider deepseek \
-            --model deepseek-v4-pro \
-            --tools "$PI_TOOLS" \
-            "${PI_HARDEN_FLAGS[@]}" \
-            --mode json \
-            < /tmp/pi-prompt.md \
-            | tee "$OUT_DIR/pi-events.jsonl" \
-            | jq -rc --unbuffered '
-                if .type == "agent_start" then "🤖 pi agent started"
-                elif .type == "turn_start" then "── turn ──"
-                elif .type == "message_end" then
-                  "[\(.message.role)] " + (
-                    (.message.content // [])
-                    | map(
-                        if .type == "text" then "text(\(.text | length)c)"
-                        elif .type == "tool_use" then "🔧 \(.name) \(.input | @json | .[:160])"
-                        elif .type == "tool_result" then "✅ result"
-                        else .type
-                        end
-                      )
-                    | join(" | ")
-                  )
-                elif .type == "turn_end" then "── turn done (\((.toolResults // []) | length) tool result(s)) ──"
-                elif .type == "agent_end" then "🏁 pi agent done"
-                else empty
-                end
-              '
-
-          jq -r '
-            select(.type == "agent_end")
-            | .messages
-            | map(select(.role == "assistant"))
-            | last
-            | (.content[]? | select(.type == "text") | .text)
-          ' "$OUT_DIR/pi-events.jsonl" > "$OUT_DIR/pi-final-message.md"
-
-      - name: Post Pi review comment
-        if: steps.pi_config.outputs.enabled == 'true' && steps.pr.outputs.skip != 'true'
-        uses: actions/github-script@v7
-        env:
-          PR_NUMBER: ${{ steps.pr.outputs.pr_number }}
-          DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}
-          GH_JOB_TOKEN: ${{ github.token }}
-        with:
-          github-token: ${{ github.token }}
-          script: |
-            const fs = require('fs');
-            const path = `${process.env.RUNNER_TEMP}/pi-final-message.md`;
-            if (!fs.existsSync(path)) {
-              core.info('Pi did not produce a final message; skipping PR comment.');
-              return;
-            }
-            let body = fs.readFileSync(path, 'utf8').trim();
-            if (!body) {
-              core.info('Pi final message was empty; skipping PR comment.');
-              return;
-            }
-            // Defense-in-depth for fork reviews: the model call needs the provider
-            // credential in the environment (readable via /proc/self/environ), and
-            // the posted comment is an exfiltration channel that bypasses GitHub
-            // Actions log masking. Strip the credential if it leaked into the text.
-            for (const s of [process.env.DEEPSEEK_API_KEY, process.env.GH_JOB_TOKEN]) {
-              if (typeof s === 'string' && s.length >= 8) {
-                body = body.split(s).join('[REDACTED]');
-              }
-            }
-            body = body.trim();
-            if (!body) {
-              core.info('Pi final message was empty after redaction; skipping PR comment.');
-              return;
-            }
-            await github.rest.issues.createComment({
-              owner: context.repo.owner,
-              repo: context.repo.repo,
-              issue_number: Number(process.env.PR_NUMBER),
-              body,
-            });

.github/workflows/pr-ready-review.yml

@@ -3,200 +3,44 @@ name: Claude Auto Review
 on:
   pull_request:
     types: [ready_for_review, opened]
-  workflow_call:
-    inputs:
-      pr_number:
-        description: 'PR number to review'
-        required: true
-        type: number
-      extra_prompt:
-        description: 'Additional reviewer instructions appended to the standard review prompt'
-        required: false
-        type: string
-        default: ''
-      triggered_by:
-        description: 'GitHub username that triggered this review (for audit only)'
-        required: false
-        type: string
-        default: ''
-    secrets:
-      CLAUDE_CODE_OAUTH_TOKEN:
-        required: true
-      WINDMILL_EE_PRIVATE_ACCESS:
-        required: false
 
 concurrency:
-  group: claude-review-${{ inputs.pr_number || github.event.pull_request.number }}
+  group: claude-review-${{ github.event.pull_request.number }}
   cancel-in-progress: true
 
 jobs:
   auto-review:
     runs-on: ubuntu-latest
-    # A non-fork PR (head.repo.fork == false) can only be opened by someone with push
-    # access to this repo, so fork==false already enforces write access. Do NOT re-add
-    # an author_association gate: the pull_request webhook payload reports private org
-    # members as CONTRIBUTOR/NONE (only public members show as MEMBER), which silently
-    # skips auto-review for every private member.
-    if: |
-      github.event_name == 'workflow_call' ||
-      (
-        (github.event.pull_request.draft == false || github.event.pull_request.ready_for_review == true) &&
-        github.event.pull_request.head.repo.fork == false
-      )
+    if: github.event.pull_request.draft == false || github.event.pull_request.ready_for_review == true
     permissions:
       contents: read
       pull-requests: read
       id-token: write
     steps:
-      # An agent-driven PR flips to ready only after a clean /review round on a
-      # draft, marked by an author comment naming the head SHA (pr skill, "Review
-      # rounds"). Re-reviewing that same head on ready_for_review is redundant.
-      # The marker alone is author attestation, so also require reviewer evidence:
-      # a Codex review (posted by github-actions[bot], not forgeable by the author)
-      # that predates the marker and carries a non-blocking verdict. Comment-
-      # triggered (workflow_call) and opened runs never skip. Keep the three
-      # copies of this check in sync (pr-ready-review / codex-pr-review /
-      # pi-pr-review); a shared local action would need the repo checked out
-      # before the check, which the codex/pi fork paths deliberately avoid.
-      - name: Check clean-round marker
-        id: marker
-        if: github.event_name == 'pull_request' && github.event.action == 'ready_for_review'
-        env:
-          GH_TOKEN: ${{ github.token }}
-          REPO: ${{ github.repository }}
-          PR_NUMBER: ${{ github.event.pull_request.number }}
-          HEAD_SHA: ${{ github.event.pull_request.head.sha }}
-          PR_AUTHOR: ${{ github.event.pull_request.user.login }}
-        run: |
-          # Fetch failures fail open (skip=false): an API hiccup must run the
-          # review, never skip it or fail the job.
-          COMMENTS=$(gh api "repos/$REPO/issues/$PR_NUMBER/comments?per_page=100" --paginate | jq -s '[.[][]]') || COMMENTS='[]'
-          MARKER_TIME=$(jq -r --arg author "$PR_AUTHOR" --arg marker "✅ Review round clean @ $HEAD_SHA" \
-            '[.[] | select(.user.login == $author) | select(.body | contains($marker)) | .created_at] | min // empty' <<<"$COMMENTS")
-          CODEX_VERDICT=''
-          if [ -n "$MARKER_TIME" ]; then
-            # Only Codex evidence that predates the marker counts: the ready-
-            # triggered Codex run itself posts after the flip and must not vouch
-            # for a sibling reviewer's skip.
-            CODEX_VERDICT=$(jq -r --arg mt "$MARKER_TIME" \
-              '[.[] | select(.user.login == "github-actions[bot]") | select(.body | contains("## Codex Review")) | select(.created_at < $mt)] | last | .body // ""' <<<"$COMMENTS" \
-              | grep -m1 -oE '(Good to merge|Mergeable, but should ideally address nits|Should address issues before merging)' || true)
-          fi
-          if [ -n "$MARKER_TIME" ] && [ -n "$CODEX_VERDICT" ] && [ "$CODEX_VERDICT" != "Should address issues before merging" ]; then
-            echo "Clean review round marker found for $HEAD_SHA with pre-marker non-blocking Codex verdict; skipping redundant review."
-            echo "skip=true" >> "$GITHUB_OUTPUT"
-          else
-            echo "skip=false" >> "$GITHUB_OUTPUT"
-          fi
-
       - name: Checkout repository
-        if: steps.marker.outputs.skip != 'true'
-        uses: actions/checkout@v5
+        uses: actions/checkout@v4
         with:
           fetch-depth: 1
 
-      - name: Check EE access
-        if: steps.marker.outputs.skip != 'true'
-        id: ee
-        env:
-          EE_TOKEN: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-        run: |
-          if [ -n "$EE_TOKEN" ]; then
-            echo "available=true" >> "$GITHUB_OUTPUT"
-            echo "ee_repo_ref=$(cat ./backend/ee-repo-ref.txt)" >> "$GITHUB_OUTPUT"
-          else
-            echo "available=false" >> "$GITHUB_OUTPUT"
-          fi
-
-      - name: Checkout EE repository
-        if: steps.ee.outputs.available == 'true'
-        uses: actions/checkout@v5
-        with:
-          repository: windmill-labs/windmill-ee-private
-          path: ./windmill-ee-private
-          ref: ${{ steps.ee.outputs.ee_repo_ref }}
-          token: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-          fetch-depth: 1
-
-      - name: Substitute EE code
-        if: steps.ee.outputs.available == 'true'
-        run: ./backend/substitute_ee_code.sh --copy --dir ./windmill-ee-private
-
-      - name: Resolve PR number
-        if: steps.marker.outputs.skip != 'true'
-        id: resolve
-        env:
-          GH_TOKEN: ${{ github.token }}
-          REPO: ${{ github.repository }}
-          INPUT_PR_NUMBER: ${{ inputs.pr_number }}
-          EVENT_PR_NUMBER: ${{ github.event.pull_request.number }}
-          EVENT_PR_AUTHOR: ${{ github.event.pull_request.user.login }}
-        run: |
-          if [ -n "$INPUT_PR_NUMBER" ]; then
-            PR_NUMBER="$INPUT_PR_NUMBER"
-            PR_AUTHOR=$(gh api "repos/$REPO/pulls/$PR_NUMBER" --jq '.user.login')
-          else
-            PR_NUMBER="$EVENT_PR_NUMBER"
-            PR_AUTHOR="$EVENT_PR_AUTHOR"
-          fi
-          echo "pr_number=$PR_NUMBER" >> "$GITHUB_OUTPUT"
-          echo "pr_author=$PR_AUTHOR" >> "$GITHUB_OUTPUT"
-
-      - name: Fetch prior PR discussion
-        if: steps.marker.outputs.skip != 'true'
-        id: prior
-        env:
-          GH_TOKEN: ${{ github.token }}
-          REPO: ${{ github.repository }}
-          PR_NUMBER: ${{ steps.resolve.outputs.pr_number }}
-        run: |
-          gh api "repos/$REPO/issues/$PR_NUMBER/comments?per_page=100" \
-            --jq '[.[] | {user: .user.login, created_at: .created_at, body: (.body | .[:4000])}] | sort_by(.created_at) | .[-20:]' \
-            > prior-comments.json || echo "[]" > prior-comments.json
-          jq -r '
-            if length == 0 then ""
-            else
-              "## Prior PR discussion (most recent up to 20 comments)\n\nIf you have already reviewed this PR (look for your own earlier comment), focus on what changed since then per the diff and respect any decisions the human made in replies. Do not re-flag findings the human already pushed back on.\n\n" +
-              (map("### @\(.user) (\(.created_at))\n\n\(.body)") | join("\n\n---\n\n"))
-            end
-          ' prior-comments.json > prior-comments.md
-
       - name: Read review prompt
-        if: steps.marker.outputs.skip != 'true'
         id: review-prompt
-        env:
-          EXTRA_PROMPT: ${{ inputs.extra_prompt }}
         run: |
           {
             echo 'REVIEW_PROMPT<<EOF'
-            cat REVIEW.md
-            echo ''
             cat .claude/review-prompt.md
-            if [ -n "$EXTRA_PROMPT" ]; then
-              echo ''
-              echo '## Additional reviewer instructions'
-              echo ''
-              printf '%s\n' "$EXTRA_PROMPT"
-            fi
-            if [ -s prior-comments.md ]; then
-              echo ''
-              cat prior-comments.md
-            fi
             echo 'EOF'
           } >> "$GITHUB_ENV"
 
       - name: Automatic PR Review
-        if: steps.marker.outputs.skip != 'true'
         uses: anthropics/claude-code-action@v1
         with:
           claude_code_oauth_token: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
           track_progress: true
           prompt: |
             REPO: ${{ github.repository }}
-            PR NUMBER: ${{ steps.resolve.outputs.pr_number }}
-            PR AUTHOR: ${{ steps.resolve.outputs.pr_author }}
+            PR NUMBER: ${{ github.event.pull_request.number }}
 
             ${{ env.REVIEW_PROMPT }}
           claude_args: |
             --allowedTools "mcp__github_inline_comment__create_inline_comment,Bash(gh pr comment:*),Bash(gh pr diff:*),Bash(gh pr view:*)"
-            --model claude-opus-5
+            --model opus

.github/workflows/pr-review-commands.yml

@@ -1,310 +0,0 @@
-name: PR Review Commands
-
-on:
-  issue_comment:
-    types: [created]
-
-jobs:
-  parse:
-    if: github.event.issue.pull_request != null
-    runs-on: ubuntu-latest
-    outputs:
-      command: ${{ steps.parse.outputs.command }}
-      extra_prompt: ${{ steps.parse.outputs.extra_prompt }}
-    steps:
-      - name: Parse command from comment
-        id: parse
-        env:
-          BODY: ${{ github.event.comment.body }}
-        run: |
-          FIRST_LINE=$(printf '%s' "$BODY" | head -n 1 | sed -E 's/^[[:space:]]+//; s/[[:space:]]+$//')
-          FIRST_WORD=${FIRST_LINE%% *}
-          case "$FIRST_WORD" in
-            /review|/codex|/pi|/claude)
-              COMMAND="${FIRST_WORD#/}"
-              REMAINDER_FIRST_LINE=${FIRST_LINE#"$FIRST_WORD"}
-              REMAINDER_FIRST_LINE=${REMAINDER_FIRST_LINE# }
-              REST=$(printf '%s' "$BODY" | tail -n +2)
-              {
-                echo "command=$COMMAND"
-                echo 'extra_prompt<<EXTRA_EOF'
-                if [ -n "$REMAINDER_FIRST_LINE" ]; then
-                  printf '%s\n' "$REMAINDER_FIRST_LINE"
-                fi
-                if [ -n "$REST" ]; then
-                  printf '%s\n' "$REST"
-                fi
-                echo 'EXTRA_EOF'
-              } >> "$GITHUB_OUTPUT"
-              ;;
-            *)
-              echo "command=" >> "$GITHUB_OUTPUT"
-              ;;
-          esac
-
-  # author_association misses private org members; check-access resolves them via the
-  # internal app token. Both are OR'd so public members still pass instantly.
-  check-access:
-    needs: [parse]
-    if: needs.parse.outputs.command != ''
-    uses: ./.github/workflows/check-write-access.yml
-    with:
-      username: ${{ github.event.comment.user.login }}
-    secrets: inherit
-
-  acknowledge:
-    needs: [parse, check-access]
-    if: |
-      needs.parse.outputs.command != '' &&
-      (
-        contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association) ||
-        needs.check-access.outputs.authorized == 'true'
-      )
-    runs-on: ubuntu-latest
-    permissions:
-      issues: write
-      pull-requests: write
-    steps:
-      - name: React to comment with eyes
-        env:
-          GH_TOKEN: ${{ github.token }}
-          REPO: ${{ github.repository }}
-          COMMENT_ID: ${{ github.event.comment.id }}
-        run: |
-          gh api -X POST \
-            "/repos/$REPO/issues/comments/$COMMENT_ID/reactions" \
-            -f content=eyes >/dev/null
-
-  # Decide, per agent, whether to launch a fresh run, re-run in place, or skip. A push
-  # already auto-triggers codex/pi (and claude on open) against the PR head. Relaunching
-  # via this issue_comment path both cancels those in-flight auto runs (shared concurrency
-  # group) AND lands the new run's status on main — issue_comment runs never attach a
-  # check to the PR head — leaving the PR showing only a cancelled review. So for every
-  # command, launch an agent only when nothing covers the head commit; if the head's run
-  # was cancelled/failed, re-run it in place (a re-run keeps the original pull_request
-  # event, so its checks re-attach to the PR head); skip when a running or successful run
-  # already covers it. `/review` applies this to all three agents; `/codex`, `/pi`,
-  # `/claude` apply the same decision to just their own agent.
-  plan:
-    needs: [parse, check-access]
-    if: |
-      needs.parse.outputs.command != '' &&
-      (
-        contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association) ||
-        needs.check-access.outputs.authorized == 'true'
-      )
-    runs-on: ubuntu-latest
-    permissions:
-      contents: read
-      actions: write
-      pull-requests: read
-      statuses: write
-    outputs:
-      head_sha: ${{ steps.plan.outputs.head_sha }}
-      launch_codex: ${{ steps.plan.outputs.launch_codex }}
-      launch_pi: ${{ steps.plan.outputs.launch_pi }}
-      launch_claude: ${{ steps.plan.outputs.launch_claude }}
-    steps:
-      - name: Decide per-agent launch vs re-run for the head commit
-        id: plan
-        env:
-          GH_TOKEN: ${{ github.token }}
-          REPO: ${{ github.repository }}
-          PR_NUMBER: ${{ github.event.issue.number }}
-          COMMAND: ${{ needs.parse.outputs.command }}
-        run: |
-          set -euo pipefail
-
-          HEAD_SHA=$(gh pr view "$PR_NUMBER" --repo "$REPO" --json headRefOid --jq '.headRefOid')
-          echo "PR #$PR_NUMBER head: $HEAD_SHA"
-          echo "head_sha=$HEAD_SHA" >> "$GITHUB_OUTPUT"
-
-          RUN_URL="$GITHUB_SERVER_URL/$REPO/actions/runs/$GITHUB_RUN_ID"
-
-          # A fresh launch runs from this issue_comment workflow (associated with main),
-          # so it never appears in the PR-head run query below and its own check lands on
-          # main, not the head. To keep fresh launches idempotent per head, mark the head
-          # SHA with a `review-launch/<agent>` commit status at launch; the `finalize` job
-          # resolves it to success/failure. A prior launch's status covering the head lets
-          # a second comment skip instead of relaunching (which would cancel the first via
-          # the reviewer's shared concurrency group). All status calls are best-effort — a
-          # GitHub API hiccup must degrade to a relaunch, never abort the decision.
-          mark_launch() {
-            agent="$1"
-            gh api -X POST "repos/$REPO/statuses/$HEAD_SHA" \
-              -f state=pending -f "context=review-launch/$agent" -f "target_url=$RUN_URL" \
-              -f "description=Review launched via /$COMMAND" >/dev/null 2>&1 || true
-          }
-
-          # Returns "covered" if a prior fresh launch (this or an earlier comment run)
-          # already covers the head: a success status, or a pending status whose launching
-          # run is still alive. A pending whose run has completed is stale (that run
-          # crashed before finalize) and does not count.
-          launch_coverage() {
-            agent="$1"
-            st_json=$(gh api "repos/$REPO/commits/$HEAD_SHA/statuses" \
-              --jq "[.[] | select(.context == \"review-launch/$agent\")] | first // empty" 2>/dev/null || true)
-            [ -n "$st_json" ] || return 0
-            state=$(jq -r '.state // empty' <<<"$st_json" 2>/dev/null || true)
-            [ "$state" = success ] && { echo covered; return 0; }
-            [ "$state" = pending ] || return 0
-            target=$(jq -r '.target_url // empty' <<<"$st_json" 2>/dev/null || true)
-            run_id=$(printf '%s' "$target" | grep -oE '[0-9]+$' || true)
-            if [ -n "$run_id" ]; then
-              run_state=$(gh run view "$run_id" --repo "$REPO" --json status --jq '.status' 2>/dev/null || true)
-              [ "$run_state" = completed ] && return 0   # stale pending -> not covered
-            fi
-            echo covered
-          }
-
-          decide() {
-            wf="$1"; key="$2"; agent="$3"
-            if [ "$(launch_coverage "$agent")" = covered ]; then
-              echo "$key: a prior launch already covers $HEAD_SHA (review-launch/$agent) -> skip"
-              echo "$key=false" >> "$GITHUB_OUTPUT"
-              return
-            fi
-            # `--commit` matches runs whose head SHA is the PR head. Auto reviews run on
-            # `pull_request` against that SHA; `/review` (issue_comment) runs execute on
-            # main, so they never match and are not counted as covering the head commit.
-            runs=$(gh run list --repo "$REPO" --workflow "$wf" --commit "$HEAD_SHA" --limit 40 \
-              --json databaseId,status,conclusion)
-            # Healthy = still running, or completed successfully: a review already
-            # covers this commit, so skip.
-            healthy=$(jq -r '[.[] | select(.status != "completed" or .conclusion == "success")] | length' <<<"$runs")
-            if [ "$healthy" -gt 0 ]; then
-              echo "$key: a running or successful review already covers $HEAD_SHA -> skip"
-              echo "$key=false" >> "$GITHUB_OUTPUT"
-              return
-            fi
-            # Re-run only genuinely interrupted runs (cancelled/failed/timed out) in
-            # place, so their checks re-attach to the PR head instead of posting on
-            # main. A `skipped` run produced no review and would just skip again (it is
-            # the draft/fork gate), so it does not count — fall through to a fresh launch.
-            retry_id=$(jq -r '[.[] | select(.status == "completed" and (.conclusion == "cancelled" or .conclusion == "failure" or .conclusion == "timed_out"))] | sort_by(.databaseId) | last | .databaseId // empty' <<<"$runs")
-            if [ -n "$retry_id" ]; then
-              if gh run rerun "$retry_id" --repo "$REPO" >/dev/null 2>&1; then
-                echo "$key: re-ran interrupted run $retry_id (re-attaches to PR head)"
-                echo "$key=false" >> "$GITHUB_OUTPUT"
-                return
-              fi
-              echo "$key: re-run of $retry_id failed -> fresh launch"
-              mark_launch "$agent"
-              echo "$key=true" >> "$GITHUB_OUTPUT"
-              return
-            fi
-            echo "$key: no usable review for $HEAD_SHA -> launch"
-            mark_launch "$agent"
-            echo "$key=true" >> "$GITHUB_OUTPUT"
-          }
-
-          # `/review` targets all three agents; `/codex`, `/pi`, `/claude` target only
-          # their own. A non-targeted agent is left untouched (no launch, no re-run).
-          decide_if_targeted() {
-            wf="$1"; key="$2"; agent="$3"
-            if [ "$COMMAND" = review ] || [ "$COMMAND" = "$agent" ]; then
-              decide "$wf" "$key" "$agent"
-            else
-              echo "$key: /$COMMAND does not target $agent -> skip"
-              echo "$key=false" >> "$GITHUB_OUTPUT"
-            fi
-          }
-
-          decide_if_targeted codex-pr-review.yml launch_codex codex
-          decide_if_targeted pi-pr-review.yml launch_pi pi
-          decide_if_targeted pr-ready-review.yml launch_claude claude
-
-  claude:
-    needs: [parse, check-access, plan]
-    if: |
-      (
-        contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association) ||
-        needs.check-access.outputs.authorized == 'true'
-      ) &&
-      needs.plan.outputs.launch_claude == 'true'
-    permissions:
-      contents: read
-      pull-requests: read
-      id-token: write
-    uses: ./.github/workflows/pr-ready-review.yml
-    with:
-      pr_number: ${{ github.event.issue.number }}
-      extra_prompt: ${{ needs.parse.outputs.extra_prompt }}
-      triggered_by: ${{ github.event.comment.user.login }}
-    secrets:
-      CLAUDE_CODE_OAUTH_TOKEN: ${{ secrets.CLAUDE_CODE_OAUTH_TOKEN }}
-      WINDMILL_EE_PRIVATE_ACCESS: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-
-  codex:
-    needs: [parse, check-access, plan]
-    if: |
-      (
-        contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association) ||
-        needs.check-access.outputs.authorized == 'true'
-      ) &&
-      needs.plan.outputs.launch_codex == 'true'
-    permissions:
-      contents: read
-      issues: write
-      pull-requests: write
-    uses: ./.github/workflows/codex-pr-review.yml
-    with:
-      pr_number: ${{ github.event.issue.number }}
-      extra_prompt: ${{ needs.parse.outputs.extra_prompt }}
-      triggered_by: ${{ github.event.comment.user.login }}
-    secrets:
-      OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
-      CODEX_AUTH_JSON: ${{ secrets.CODEX_AUTH_JSON }}
-      WINDMILL_EE_PRIVATE_ACCESS: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-
-  pi:
-    needs: [parse, check-access, plan]
-    if: |
-      (
-        contains(fromJSON('["OWNER", "MEMBER", "COLLABORATOR"]'), github.event.comment.author_association) ||
-        needs.check-access.outputs.authorized == 'true'
-      ) &&
-      needs.plan.outputs.launch_pi == 'true'
-    permissions:
-      contents: read
-      issues: write
-      pull-requests: write
-    uses: ./.github/workflows/pi-pr-review.yml
-    with:
-      pr_number: ${{ github.event.issue.number }}
-      extra_prompt: ${{ needs.parse.outputs.extra_prompt }}
-      triggered_by: ${{ github.event.comment.user.login }}
-    secrets:
-      DEEPSEEK_API_KEY: ${{ secrets.DEEPSEEK_API_KEY }}
-      WINDMILL_EE_PRIVATE_ACCESS: ${{ secrets.WINDMILL_EE_PRIVATE_ACCESS }}
-
-  # Resolve the `review-launch/<agent>` head statuses that `plan` set to pending, so a
-  # fresh launch's outcome is visible on the PR head (not just on main) and never lingers
-  # as a stale pending check. Targets the exact SHA `plan` launched against, so a push
-  # that moved the head mid-review does not stamp a status on the new head.
-  finalize:
-    needs: [plan, claude, codex, pi]
-    if: always() && needs.plan.result == 'success' && needs.plan.outputs.head_sha != ''
-    runs-on: ubuntu-latest
-    permissions:
-      statuses: write
-    env:
-      GH_TOKEN: ${{ github.token }}
-      REPO: ${{ github.repository }}
-      HEAD_SHA: ${{ needs.plan.outputs.head_sha }}
-      RUN_URL: ${{ github.server_url }}/${{ github.repository }}/actions/runs/${{ github.run_id }}
-    steps:
-      - name: Finalize launch statuses on the PR head
-        run: |
-          set -uo pipefail
-          finalize() {
-            agent="$1"; launched="$2"; result="$3"
-            [ "$launched" = true ] || return 0
-            state=$([ "$result" = success ] && echo success || echo failure)
-            gh api -X POST "repos/$REPO/statuses/$HEAD_SHA" \
-              -f "state=$state" -f "context=review-launch/$agent" -f "target_url=$RUN_URL" \
-              -f "description=Review $result" >/dev/null 2>&1 || true
-          }
-          finalize codex "${{ needs.plan.outputs.launch_codex }}" "${{ needs.codex.result }}"
-          finalize pi "${{ needs.plan.outputs.launch_pi }}" "${{ needs.pi.result }}"
-          finalize claude "${{ needs.plan.outputs.launch_claude }}" "${{ needs.claude.result }}"

.github/workflows/publish-cli-docs.yml

@@ -1,84 +0,0 @@
-name: Publish CLI docs repo
-
-# Regenerates the windmill-cli-docs repo (consumed by context7) from the
-# canonical sources in this repo on every Windmill release.
-#
-# Required secret:
-#   CLI_DOCS_DEPLOY_KEY — ed25519 private key whose public half is registered
-#                        as a write-access deploy key on
-#                        windmill-labs/windmill-cli-docs.
-
-on:
-  push:
-    tags:
-      - "v*"
-  workflow_dispatch:
-
-# Serialize pushes to windmill-cli-docs so two release tags landing close
-# together (e.g. a release-please bump + a hotfix) can't race to force-push
-# the docs repo.
-concurrency:
-  group: publish-cli-docs
-  cancel-in-progress: false
-
-jobs:
-  publish:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout windmill (source of truth)
-        uses: actions/checkout@v4
-        with:
-          path: windmill
-
-      - name: Checkout windmill-cli-docs (publish target)
-        uses: actions/checkout@v4
-        with:
-          repository: windmill-labs/windmill-cli-docs
-          path: windmill-cli-docs
-          ssh-key: ${{ secrets.CLI_DOCS_DEPLOY_KEY }}
-          fetch-depth: 0
-
-      - uses: actions/setup-python@v5
-        with:
-          python-version: "3.12"
-
-      - name: Install dependencies
-        run: pip install pyyaml
-
-      - name: Regenerate docs
-        run: |
-          python3 windmill/system_prompts/generate.py \
-            --context7-dir "$GITHUB_WORKSPACE/windmill-cli-docs"
-
-      - name: Commit and push if changed
-        working-directory: windmill-cli-docs
-        env:
-          REF_NAME: ${{ github.ref_name }}
-          REF_TYPE: ${{ github.ref_type }}
-        run: |
-          git config user.name "windmill-bot"
-          git config user.email "bot@windmill.dev"
-          git add -A
-          if git diff --cached --quiet; then
-            echo "No doc changes for ${REF_NAME}."
-            committed=false
-          else
-            committed=true
-            if [ "${REF_TYPE}" = "tag" ]; then
-              git commit -m "chore: sync from windmill ${REF_NAME}"
-            else
-              git commit -m "chore: sync from windmill (manual dispatch from ${REF_NAME})"
-            fi
-            git push origin HEAD
-          fi
-          # Always mirror the version tag on tag pushes, even when content
-          # didn't change — downstream consumers tie snapshots to releases by
-          # tag, and skipping it would leave the docs repo without a tag for
-          # the new Windmill release.
-          # workflow_dispatch from a non-tag ref skips this so we don't
-          # create a junk tag named after a branch.
-          if [ "${REF_TYPE}" = "tag" ]; then
-            git tag -f "${REF_NAME}"
-            git push origin "${REF_NAME}" --force
-            echo "Mirrored tag ${REF_NAME} to windmill-cli-docs (content changed: ${committed})."
-          fi

.github/workflows/publish_windows_worker.yml

@@ -35,7 +35,7 @@ jobs:
       - uses: actions-rust-lang/setup-rust-toolchain@v1
         with:
           cache-workspaces: backend
-          toolchain: 1.97.0
+          toolchain: 1.93.0
 
       - name: Substitute EE code
         shell: bash
@@ -56,12 +56,8 @@ jobs:
           vcpkg.exe integrate install
           $env:VCPKGRS_DYNAMIC=1
           $env:OPENSSL_DIR="${Env:VCPKG_INSTALLATION_ROOT}\installed\x64-windows-static"
-          cd backend
-          # Stub the openapi specs to empty: they are compiled in via an ungated
-          # include_str! but a worker binary never serves them, so this avoids
-          # embedding ~2.5MB of spec.
+          mkdir frontend/build && cd backend
           New-Item -Path . -Name "windmill-api/openapi-deref.yaml" -ItemType "File" -Force
-          New-Item -Path . -Name "windmill-api/openapi-deref.json" -ItemType "File" -Force
           cargo build --release --features=ee_windows
       - name: Rename binary with corresponding architecture
         run: |

.github/workflows/refresh-docs-snapshot.yml

@@ -1,52 +0,0 @@
-name: Refresh docs snapshot
-
-# The backend embeds a vendored docs snapshot (backend/windmill-api/docs_snapshot/*.gz)
-# so in-product docs search works with no runtime egress. This job re-fetches it from
-# windmill.dev on a schedule and opens a PR when it changed, keeping the embedded docs
-# fresh independently of the release cadence (the binary embeds whatever is on the
-# source tree at build time, so a merged refresh rides into the next release build).
-on:
-  schedule:
-    - cron: "0 6 * * 1" # Mondays 06:00 UTC
-  workflow_dispatch:
-
-jobs:
-  refresh:
-    runs-on: ubicloud
-    permissions:
-      contents: write
-      pull-requests: write
-    steps:
-      - uses: actions/create-github-app-token@v2
-        id: app
-        with:
-          app-id: ${{ vars.INTERNAL_APP_ID }}
-          private-key: ${{ secrets.INTERNAL_APP_KEY }}
-      - uses: actions/checkout@v4
-        with:
-          token: ${{ steps.app.outputs.token }}
-      - name: Fetch + re-gzip docs snapshot
-        run: cd backend/windmill-api/docs_snapshot && ./fetch.sh
-      - name: Sanity-check the fetched corpus
-        # curl -f in fetch.sh rejects HTTP errors, but not a valid-but-garbage 200
-        # (truncated file, error page). Guard against embedding a broken snapshot.
-        run: |
-          cd backend/windmill-api/docs_snapshot
-          test "$(wc -c < llms-full.txt.gz)" -gt 100000
-          test "$(wc -c < llms.txt.gz)" -gt 1000
-          pages=$(gzip -dc llms-full.txt.gz | grep -c '^Source:' || true)
-          echo "pages in snapshot: $pages"
-          test "${pages:-0}" -ge 200
-      - uses: peter-evans/create-pull-request@v6
-        with:
-          token: ${{ steps.app.outputs.token }}
-          branch: chore/refresh-docs-snapshot
-          add-paths: backend/windmill-api/docs_snapshot/*.gz
-          commit-message: "chore: refresh vendored docs snapshot"
-          title: "chore: refresh vendored docs snapshot"
-          body: |
-            Automated refresh of the embedded docs snapshot
-            (`backend/windmill-api/docs_snapshot/*.gz`) from windmill.dev.
-
-            Review the diff for unexpected churn (a bad upstream docs deploy would
-            show up as a large drop in pages or content) before merging.

.github/workflows/release-please.yml

@@ -8,10 +8,8 @@ jobs:
     name: "Release please"
     runs-on: ubicloud
     steps:
-      # Config lives in release-please-config.json / .release-please-manifest.json:
-      # a `release-type` input instead re-derives the last released version by
-      # paginating every GitHub release, which on a repo this size is slow enough
-      # to fail intermittently.
-      - uses: googleapis/release-please-action@v5
+      - uses: GoogleCloudPlatform/release-please-action@v3
         with:
+          release-type: simple
+          package-name: windmill
           token: ${{ secrets.PAT_TOKEN }}

.github/workflows/sdk-tests.yml

@@ -1,61 +0,0 @@
-# The python and typescript SDK unit suites, on release tags only: they guard
-# what gets published to npm / PyPI / JSR, and a tag is the moment that decides
-# it.
-#
-# This runs alongside the publish workflows rather than ahead of them, so it
-# reports a broken SDK rather than holding one back. Gating would mean putting
-# the job inside each publish workflow, since Actions cannot express `needs`
-# across workflows.
-name: SDK Tests
-
-on:
-  workflow_dispatch:
-  push:
-    tags:
-      - "v*"
-
-jobs:
-  typescript-client:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Setup Bun
-        uses: oven-sh/setup-bun@v2
-        with:
-          bun-version: latest
-
-      # No build step: these suites are deliberately free of the generated API
-      # client, so they run against the sources as committed.
-      - name: Run tests
-        working-directory: ./typescript-client
-        run: bun test --timeout 120000 tests/
-
-  python-client:
-    runs-on: ubuntu-latest
-    steps:
-      - name: Checkout code
-        uses: actions/checkout@v4
-
-      - name: Setup uv
-        uses: astral-sh/setup-uv@v5
-
-      # The interpreter is named explicitly: on a clean checkout uv picks the
-      # runner's system python and stops with "not compatible with the locked
-      # Python requirement" rather than fetching one. Keep in step with
-      # `requires-python` in uv.lock.
-      #
-      # Note this is not the version a worker runs the SDK on — those are 3.12.
-      # `uv.lock` asks for >=3.14, so pinning lower means regenerating it, which
-      # is worth doing separately.
-      - name: Install the interpreter the lockfile requires
-        run: uv python install 3.14
-
-      # `--frozen` so a drifted lockfile fails here rather than quietly
-      # resolving to something nobody has run.
-      - name: Run tests
-        working-directory: ./python-client/wmill
-        env:
-          PYTHONPATH: .
-        run: uv run --frozen --python 3.14 pytest tests/ -q

.github/workflows/spawn-ephemeral-backend.yml

@@ -0,0 +1,126 @@
+name: Spawn Ephemeral Backend
+
+on:
+  issue_comment:
+    types: [created]
+  pull_request_review_comment:
+    types: [created]
+  workflow_dispatch:
+    inputs:
+      pr_number:
+        description: "PR number"
+        required: true
+        type: number
+
+jobs:
+  check-membership:
+    if: |
+      (github.event_name == 'issue_comment' && contains(github.event.comment.body, '/spawnbackend')) ||
+      (github.event_name == 'pull_request_review_comment' && contains(github.event.comment.body, '/spawnbackend'))
+    uses: ./.github/workflows/check-org-membership.yml
+    secrets:
+      access_token: ${{ secrets.ORG_ACCESS_TOKEN }}
+
+  spawn-backend:
+    needs: check-membership
+    # Only run on PR comments that contain /spawn-backend, or manual dispatch
+    if: |
+      github.event_name == 'workflow_dispatch' ||
+      (github.event.issue.pull_request && needs.check-membership.outputs.is_member == 'true')
+    runs-on: ubuntu-latest
+    permissions:
+      pull-requests: write
+      contents: read
+
+    steps:
+      - name: Get PR details
+        id: pr-details
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const prNumber = context.eventName === 'workflow_dispatch'
+              ? context.payload.inputs.pr_number
+              : context.issue.number;
+
+            const pr = await github.rest.pulls.get({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              pull_number: prNumber
+            });
+
+            // Get branch name and format it for Cloudflare Pages
+            // Replace '/' with '-' for the URL
+            const branchName = pr.data.head.ref;
+            const formattedBranch = branchName.replace(/\//g, '-');
+            const cfFrontendUrl = `https://${formattedBranch}.windmill.pages.dev`;
+
+            core.setOutput('commit_hash', pr.data.head.sha);
+            core.setOutput('pr_number', prNumber);
+            core.setOutput('branch_name', branchName);
+            core.setOutput('cf_frontend_url', cfFrontendUrl);
+
+      - name: Check manager URL
+        id: check-manager-url
+        run: |
+          if [ -z "${{ secrets.EPHEMERAL_BACKEND_QUEUE_URL }}" ]; then
+            echo "manager_url_set=false" >> $GITHUB_OUTPUT
+          else
+            echo "manager_url_set=true" >> $GITHUB_OUTPUT
+          fi
+
+      - name: Post error comment if manager not running
+        if: steps.check-manager-url.outputs.manager_url_set == 'false'
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const prNumber = context.eventName === 'workflow_dispatch'
+              ? Number(context.payload.inputs.pr_number)
+              : context.issue.number;
+
+            await github.rest.issues.createComment({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: prNumber,
+              body: `❌ Manager URL not set (did you start the ephemeral backend manager?)\n\nThe ephemeral backend manager needs to be running to spawn backends. Please start the manager first.`
+            });
+
+      - name: Fail if manager not running
+        if: steps.check-manager-url.outputs.manager_url_set == 'false'
+        run: |
+          echo "Error: EPHEMERAL_BACKEND_QUEUE_URL secret is not set"
+          exit 1
+
+      - name: Trigger Windmill flow
+        if: steps.check-manager-url.outputs.manager_url_set == 'true'
+        id: trigger-flow
+        run: |
+          JOB_UUID=$(curl -s -X POST "https://app.windmill.dev/api/w/windmill-labs/jobs/run/f/f/all/run_ephemeral_backend" \
+            -H "Authorization: Bearer ${{ secrets.WINDMILL_RUN_FLOW_TOKEN }}" \
+            -H "Content-Type: application/json" \
+            -d '{
+              "manager_url": "${{ secrets.EPHEMERAL_BACKEND_QUEUE_URL }}",
+              "commit_hash": "${{ steps.pr-details.outputs.commit_hash }}",
+              "pr_number": ${{ steps.pr-details.outputs.pr_number }},
+              "cf_frontend_url": "${{ steps.pr-details.outputs.cf_frontend_url }}"
+            }' | tr -d '"')
+
+          echo "Job UUID: $JOB_UUID"
+          echo "job_uuid=$JOB_UUID" >> $GITHUB_OUTPUT
+
+      - name: Post comment with job link
+        if: steps.check-manager-url.outputs.manager_url_set == 'true'
+        uses: actions/github-script@v7
+        with:
+          script: |
+            const jobUuid = '${{ steps.trigger-flow.outputs.job_uuid }}';
+            const appUrl = `https://app.windmill.dev/public/windmill-labs/a106bad0256c1dfa7a4f9279c42b1a4b#${jobUuid}`;
+            const prNumber = context.eventName === 'workflow_dispatch'
+              ? Number(context.payload.inputs.pr_number)
+              : context.issue.number;
+
+            await github.rest.issues.createComment({
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: prNumber,
+              body: `🚀 Spawning new ephemeral backend!\n\n${appUrl}`
+            });

.github/workflows/yaml-validator-tests.yml

@@ -1,37 +0,0 @@
-name: YAML validator tests
-
-# The schemas behind `wmill lint` are generated from the OpenAPI specs, so a spec change
-# can turn a valid synced file into a lint error without touching any validator code.
-
-on:
-  push:
-    branches: [main]
-    paths:
-      - "windmill-yaml-validator/**"
-      - "openflow.openapi.yaml"
-      - "backend/windmill-api/openapi.yaml"
-      - ".github/workflows/yaml-validator-tests.yml"
-  pull_request:
-    paths:
-      - "windmill-yaml-validator/**"
-      - "openflow.openapi.yaml"
-      - "backend/windmill-api/openapi.yaml"
-      - ".github/workflows/yaml-validator-tests.yml"
-
-jobs:
-  test:
-    runs-on: ubuntu-latest
-    steps:
-      - uses: actions/checkout@v4
-
-      - uses: actions/setup-node@v4
-        with:
-          node-version: "20"
-
-      - name: Install dependencies
-        working-directory: windmill-yaml-validator
-        run: npm ci
-
-      - name: Run tests
-        working-directory: windmill-yaml-validator
-        run: npm test

.gitignore

@@ -20,19 +20,13 @@ rust-client/Cargo.toml
 
 # Worktree-specific Claude Code settings (generated by scripts/worktree-env)
 .claude/settings.local.json
-.claude/worktrees/
 
 # Symlinked cache directories (for git worktrees)
 backend/target
-node_modules/
 frontend/node_modules
 typescript-client/node_modules
-ai_evals/node_modules
-ai_evals/results/
 frontend/.svelte-kit
 backend/chrome_profiler.json
 .fast-check/
 __pycache__/
 .playwright-mcp/
-.codex
-.claude/scheduled_tasks.lock

.mcp.json

@@ -3,16 +3,6 @@
     "svelte": {
       "type": "http",
       "url": "https://mcp.svelte.dev/mcp"
-    },
-    "playwright": {
-      "type": "stdio",
-      "command": "npx",
-      "args": ["-y", "@playwright/mcp@latest", "--browser", "chromium", "--headless", "--output-dir", "/tmp/playwright-mcp-${USER:-shared}"]
-    },
-    "playwright-headed": {
-      "type": "stdio",
-      "command": "npx",
-      "args": ["-y", "@playwright/mcp@latest", "--browser", "chromium", "--output-dir", "/tmp/playwright-mcp-${USER:-shared}"]
     }
   }
 }

.release-please-manifest.json

@@ -1,3 +0,0 @@
-{
-  ".": "1.784.0"
-}

.webmux.yaml

@@ -47,7 +47,6 @@ profiles:
       For this window specifically, backend is running on: ${BACKEND_PORT} and frontend is running on: ${FRONTEND_PORT}.
       To connect to the database, use this connection string: ${DATABASE_URL}
       Because we are running backend with cargo watch, to verify your changes, just check the logs in the backend pane. No need for cargo check.
-      For UI verification, use the Playwright MCP (`mcp__playwright__*`) — the `playwright` server is headless and works without a display. Navigate to http://localhost:${FRONTEND_PORT}, log in as admin@windmill.dev / changeme.
       IMPORTANT: Read docs/autonomous-mode.md before starting any work.
     panes:
       - id: agent
@@ -77,7 +76,6 @@ profiles:
       On this window specifically, frontend is running on: ${FRONTEND_PORT}.
       To connect to the database, use this connection string: ${DATABASE_URL}
       Because we are running frontend with npm run dev, to verify your changes, just check the logs in the frontend pane. No need for npm run build.
-      For UI verification, use the Playwright MCP (`mcp__playwright__*`) — the `playwright` server is headless and works without a display. Navigate to http://localhost:${FRONTEND_PORT}, log in as admin@windmill.dev / changeme.
       IMPORTANT: Read docs/autonomous-mode.md before starting any work.
     panes:
       - id: agent
@@ -102,55 +100,9 @@ profiles:
 
 integrations:
   github:
-    autoRemoveOnMerge: true
     linkedRepos:
       - repo: windmill-labs/windmill-ee-private
         alias: ee-private
         dir: ../windmill-ee-private__worktrees
   linear:
     enabled: true
-    autoCreateWorktrees: true
-    watchTeams: [WIN,GIT]
-
-oneshot:
-  systemPrompt: |
-    You are running in webmux ONESHOT mode.
-
-    # No interactive user
-    There is NO interactive user — nobody is watching the chat or will respond
-    to questions, approvals, or status checks. Any message asking the user to
-    review, approve, confirm, take a look, or "let you know" is wasted output:
-    it will not be answered.
-
-    # Your job
-    Take the task to its real conclusion without pausing:
-      1. Make the change.
-      2. Validate it (run the relevant tests, typecheck, build, or quick
-         manual check). For UI changes, drive the running frontend with
-         the Playwright MCP (`mcp__playwright__*`, headless) and confirm
-         the change works end-to-end before moving on.
-      3. Commit.
-      4. Push.
-      5. Open a pull request.
-    Only then are you done.
-
-    # Decisions
-    When something is ambiguous, pick the most reasonable default and proceed.
-    When you would normally ask "should I X or Y?", just pick one and continue
-    — note the choice in the PR description if it matters.
-
-    # PR readiness
-    Default to opening the PR as a draft. If you are highly confident in the
-    change — the scope is small and well-understood, validation passed
-    cleanly, and you would not change anything if a reviewer pushed back —
-    open the PR as ready-for-review directly (omit `--draft` when invoking
-    `gh pr create`, or call `gh pr ready <number>` after creation). Err on
-    the side of draft when validation was partial, the change touches
-    public APIs or shared infrastructure, or you made a non-obvious judgment
-    call.
-
-    # Ending your turn
-    Never end your turn with a question, a suggestion to "take a look", or a
-    request for approval. Stop only when the PR is open, or when you hit a
-    technical error you cannot recover from yourself (in which case clearly
-    state the blocker).

AGENTS.md

@@ -1,182 +0,0 @@
-# Windmill
-
-Open-source platform for internal tools, workflows, API integrations, background jobs, and UIs. Rust backend + Svelte 5 frontend.
-
-## Workflow
-
-1. **Understand**: Before coding, explore the codebase (see Code Navigation below). Use `outline` to understand file structure, `body` to read specific symbols, `def`/`callers`/`callees` to trace code, `Grep` to find usages. Read `docs/` for domain context.
-2. **Plan**: For non-trivial changes, use plan mode. For large features, break into reviewable stages
-3. **Execute**: Follow coding patterns from skills (`rust-backend`, `svelte-frontend`)
-4. **Validate**: After every change, run the appropriate checks per `docs/validation.md`
-
-## Documentation
-
-- **Validation**: `docs/validation.md` — what checks to run based on what you changed
-- **Unreleased SDK changes**: `docs/wac-sdk-e2e.md` — exercising a client change on a real worker
-- **Agent workers**: `docs/agent-worker-e2e.md` — building and running one locally. An agent
-  reaches the DB only through the API, so `Connection::Http` paths are never taken by a plain
-  `cargo run`; a normal build cannot start one at all.
-- **Enterprise**: `docs/enterprise.md` — EE file conventions and PR workflow
-- **Backend patterns**: use the `rust-backend` skill when writing Rust code
-- **Frontend patterns**: use the `svelte-frontend` skill when writing Svelte code. Do NOT edit svelte files unless you have read that skill.
-- **Frontend UUIDs**: do not call `crypto.randomUUID()` in frontend code. Import `randomUUID` from `$lib/utils/uuid` instead.
-- **Code review**: review the current PR or branch against the shared review policy in `REVIEW.md` (severity triage, public-surface checklist, AGENTS.md compliance, test-coverage assessment). The skill at `.agents/skills/local-review/SKILL.md` orchestrates it. All three CLIs auto-discover the same SKILL — Claude reads `.claude/skills/` (symlinked to the canonical `.agents/skills/` file), Codex and Pi read `.agents/skills/` directly. Invoke with `/local-review` in Claude Code, `$local-review` (or `/skills` selector) in Codex, or `pi --skill local-review` / `/skill:local-review` in Pi. For a Codex-driven pass that mirrors the `codex-pr-review` GitHub action against your unpushed work (committed + uncommitted) before you push, use `/local-review-codex` (`.agents/skills/local-review-codex/`) — same `REVIEW.md` policy, `gpt-5.6-sol`, `xhigh` reasoning; requires the `codex` CLI >= 0.144.1.
-- **Domain guides**: `.claude/skills/native-trigger/` and `frontend/tutorial-system-guide.mdc`
-- **Brand/UI guidelines**: `frontend/brand-guidelines.md`
-- **Domain vocabulary**: `CONTEXT.md` — the words this codebase uses for its own concepts (step, step setting, trigger step, …). Name things the way it does.
-- **CLI commands**: when adding/modifying/removing a command, subcommand, option, or description in `cli/src/commands/`, run `python system_prompts/generate.py` to refresh `system_prompts/auto-generated/` and `cli/src/guidance/skills.gen.ts`. The CLI docs the agents use to operate `wmill` are derived from the source — stale generated files give agents the wrong flags.
-- **Session recorder**: `frontend/src/lib/components/recording/` is also the recorder `wmill app dev --recording` serves, vendored into the CLI as `cli/src/commands/app/devRecorderBundle.gen.ts`. After changing `rawAppSnapshot.ts` or `rawAppRecording.svelte.ts`, run `bun run gen:dev-recorder` from `cli/` (`cli/test/dev_recorder_bundle_unit.test.ts` fails otherwise).
-
-## Dev Environment
-
-> **In a git worktree, the ports and database below are NOT the ones to use.** Each
-> worktree gets its own backend port, frontend port and Postgres database, so the
-> defaults in this section apply only to a plain single checkout. **Discover the real
-> values before running anything** — see "Per-worktree ports and database" below.
-
-- **Backend**: `cargo run` from `backend/` (API at http://localhost:8000)
-- **DuckDB local jobs**: before running DuckDB scripts locally, build the FFI shared library with `cd backend/windmill-duckdb-ffi-internal && ./build_dev.sh`. Re-run it after clean builds or when `backend/target/debug/libwindmill_duckdb_ffi_internal.*` is missing. The bundled DuckDB compile (~2min) is cached in a per-user dir shared across worktrees, so a fresh worktree reuses it and the build is near-instant.
-- **Data pipelines (DuckLake) from source**: a plain `cargo run` (even `--features quickjs`) advertises a `duckdb` worker tag but **cannot** execute DuckDB scripts and has **no** working S3 proxy (DuckLake writes 404). Build CE DuckLake with `cargo run --features quickjs,duckdb,parquet,private` (add `,python` for Python scripts, `,enterprise,license` for EE) **and** build the FFI (bullet above). See `backend/CLAUDE.md` → "Running data pipelines (DuckLake) from source" for the exact feature sets and the two feature-gate gotchas.
-- **Frontend**: `REMOTE=http://localhost:8000 npm run dev` from `frontend/` (port 3000+)
-- **DB**: `psql postgres://postgres:changeme@localhost:5432/windmill`
-- **Login**: `admin@windmill.dev` / `changeme`
-- **Instance settings**: navigate to `/#superadmin-settings`
-- **Migrations**: use `cargo sqlx migrate add -r <name>` from `backend/` to create new migrations (never generate timestamps manually)
-
-### Per-worktree ports and database
-
-A worktree's `.env` / `.env.local` (repo root) and `backend/.env` hold its own
-`DATABASE_URL` and `PORT`; the database is typically `windmill_<branch_with_underscores>`
-(branch `dbt-runtime` → `windmill_dbt_runtime`). Read them, or discover from what is
-already running:
-
-```bash
-psql postgres://postgres:changeme@localhost:5432/postgres -tAc \
-  "select datname from pg_database where datname like 'windmill%'" | grep "$(git branch --show-current | tr - _)"
-# the port the frontend actually proxies to (REMOTE of this worktree's vite):
-for p in $(pgrep -f vite); do case "$(readlink /proc/$p/cwd)" in *"$(basename "$(git rev-parse --show-toplevel)")"*)
-  tr '\0' '\n' < /proc/$p/environ | grep -E '^REMOTE=|^PORT=';; esac; done
-```
-
-Getting these wrong is not a cheap mistake:
-
-- **`DATABASE_URL` pointed at another worktree's database silently destroys the sqlx
-  cache.** `cargo run` and `cargo sqlx prepare` both compile `sqlx::query!` against the
-  **live** database, so the wrong one fails with `relation "<your_new_table>" does not
-  exist` — and `prepare` deletes the whole `.sqlx/` directory *before* it fails, leaving
-  it gutted. Always `cp -r backend/.sqlx <tmp>/sqlx_backup` first (see the `update-sqlx`
-  skill).
-- **The frontend proxies to its own worktree's backend port, not 8000.** Starting a
-  backend on the wrong port leaves the UI up but every API call 502s, which reads like an
-  application bug rather than a misconfiguration.
-- **Kill backends by pid scoped to this worktree's cwd** (`readlink /proc/<pid>/cwd`),
-  never `pkill -f target/debug/windmill` — that kills every sibling worktree's backend.
-  Beware that a `pgrep -f "<pattern>"` in a shell whose own command line contains
-  `<pattern>` matches the shell itself.
-
-## Verifying Frontend Changes
-
-After modifying frontend code, drive the running dev server with the **Playwright MCP** to verify the change in a real browser — don't claim a UI change works without exercising it.
-
-Two MCP servers are registered in `.mcp.json`:
-- `playwright` — headless Chromium, default for devboxes (no display required)
-- `playwright-headed` — windowed Chromium, when a display is available
-
-**One-time setup:** run `npx playwright install chromium` to download the browser binary (Playwright won't fetch it automatically on first use).
-
-Typical flow:
-1. Ensure backend (`cargo run`) and frontend (`REMOTE=http://localhost:8000 npm run dev`) are running
-2. `mcp__playwright__browser_navigate` to the relevant page (login at `admin@windmill.dev` / `changeme`)
-3. `mcp__playwright__browser_snapshot` to inspect the accessibility tree (preferred over screenshots for reading the DOM)
-4. `mcp__playwright__browser_click` / `browser_fill_form` / `browser_type` to interact
-5. `mcp__playwright__browser_take_screenshot` for visual confirmation
-6. `mcp__playwright__browser_console_messages` / `browser_network_requests` to surface errors
-
-Write screenshots to an absolute path under `/tmp` (the MCP servers already do; standalone
-Playwright scripts must be told): moving a PNG out of the checkout afterwards needs a `mv` the
-permission hooks always prompt on. Same reason to run `rm`/`mv`/`cp` as one plain command per Bash
-call: those hooks defer on `&&`, `;`, redirects, quotes and `$VAR`.
-
-**Attach the screenshots to the PR.** For any change under `frontend/`, embed screenshots of the affected UI in the PR body — the `pr` skill requires this and carries the upload recipe.
-
-If you cannot exercise a UI change (no dev server, etc.), say so explicitly rather than claiming success.
-
-## Verifying Backend Changes
-
-`cargo check` and the unit tests do not exercise a worker code path. **If you changed how
-a job runs — an executor, `handle_child`, anything spawning or reading from a
-subprocess — run an actual job of that kind** and confirm it completed, then say so.
-Whole classes of defect compile and unit-test clean:
-
-- **Stack overflow from a large buffer in an async block.** An array declared across an
-  `.await` is baked into the future's state; once that future is boxed a few layers deep
-  by the job poller, two 16 KB arrays abort the worker *process* (`thread
-  'tokio-runtime-worker' has overflowed its stack`). Heap-allocate read buffers
-  (`vec![0u8; N]`, not `[0u8; N]`).
-- Deadlocks from draining only one of a child's pipes, missed cancellation or timeout
-  propagation, and anything depending on the real engine's output format.
-
-A crash like this takes down every job on that worker, not just yours, so check the
-backend log after the run rather than only the job's own status. If you cannot run one,
-say which path went unexercised instead of implying it was verified.
-
-## Banned Patterns
-
-### `$bindable(default_value)` on optional props
-
-Using `$bindable(default_value)` on props that can be `undefined` is **banned**. This pattern causes subtle bugs because the default value masks the `undefined` state.
-
-**Bad:**
-
-```svelte
-let { my_prop = $bindable(default_value) }: { my_prop?: string } = $props()
-```
-
-**Correct alternatives:**
-
-1. **Use `$derived` with nullish coalescing** — handle the potential `undefined` at the usage site:
-
-   ```svelte
-   let { my_prop = $bindable() }: { my_prop?: string } = $props()
-   let effective_value = $derived(my_prop ?? default_value)
-   ```
-
-2. **Create a `useMyPropState()` helper** — encapsulate the undefined-handling logic in a reusable function and call it higher in the component tree, so the child component always receives a defined value.
-
-## Code Navigation
-
-`wm-ts-nav` is an AST-aware code navigator. Use **wm-ts-nav** for structural queries — it skips comments/strings and understands symbol boundaries.
-
-**MUST use `outline` before `Read`** on unfamiliar files — a 500-line file costs ~500 lines of context, while `outline` costs ~20. Then **MUST use `body "X"`** instead of reading a full file to see one function/struct. Use `Read` with offset/limit only when you need surrounding context that `body` doesn't capture.
-- `refs "X" --caller` instead of reading files to find which function contains each reference
-- `callers "X"` / `callees "X"` for call-graph questions
-
-EE files (`*_ee.rs`, `*_ee.ts`, `*_ee.svelte`) are indexed — you can `outline`, `def`, `body`, `refs` etc. on them just like regular files.
-
-```bash
-NAV="sh wm-ts-nav/nav"
-# Use --root backend for Rust, --root frontend/src for TS/Svelte
-$NAV --root backend outline backend/path/to/file.rs      # file structure
-$NAV --root backend def "ServiceName"                     # find definition
-$NAV --root backend body "decrypt_oauth_data"             # extract source code
-$NAV --root backend search "%" --parent ServiceName       # methods on a type
-$NAV --root backend search "Trigger" --kind struct        # find by kind
-$NAV --root backend refs "X" --file handler.rs --caller   # scoped refs with caller
-$NAV --root backend callers "X"                           # who calls X?
-$NAV --root backend callees "X"                           # what does X call?
-```
-
-**Limitations** — syntax-level analysis, no type inference. Use **Grep** instead when completeness matters (finding all usages, exhaustiveness checks):
-- `refs`/`callers`/`callees` can't follow re-exports, glob imports, or different import paths to the same symbol
-- Trait impls, macro-generated symbols (`sqlx::FromRow`), and namespace member access (`ns.X`) are invisible
-- `callees` shows all identifiers in a function body, not just actual calls
-
-## Core Principles
-
-- **MUST `outline` before `Read`** on unfamiliar files — then `body` or `Read` with offset/limit for specifics
-- Search for existing code to reuse before writing new code
-- Follow established patterns in the codebase
-- Keep changes focused — don't refactor beyond what's asked
-- **Ship only the tests the PR needs.** A committed test must pin behavior a future change could plausibly break, and be the smallest setup that exercises the new logic. While developing, write as many exhaustive tests and do as much manual testing as you need to convince yourself the change works — then remove that scaffolding before marking the PR ready, keeping only the essential regression guard(s). A test that merely re-exercises pre-existing behavior, or needs elaborate fixtures to assert something trivial, is scaffolding: delete it. If nothing meaningful is left to guard, ship no test rather than a ceremonial one.
-- **Comments record constraints, not narration.** Write a comment only for what the code can't show: why a non-obvious approach is required, what breaks if it's "simplified" away. State each invariant once, at the place where someone would break it, in ≤4 lines. Don't describe what the next line does, don't repeat the same rationale at multiple sites, and don't address the PR reviewer (justifying a change belongs in the PR description, not the code). Reference nothing ephemeral — no numbered steps from your dev flow, no "the poller / the test does X" scaffolding, no transient state that won't exist for the next reader; keep only the essential, durable rationale. Describe the code as it is, never its drafting history: "we no longer do X", "unchanged behavior", "instead of the previous approach" are meaningless to a reader who never saw the earlier iteration — before finishing, reread your comments as if the current state is the only state that ever existed.
-- **Never attribute work to a specific customer, account, or "requested by a customer" in repo-tracked content** (PR descriptions, commit messages, code comments, docs). Describe changes by their technical motivation instead.

CLAUDE.md

@@ -1 +1,87 @@
-@AGENTS.md
+# Windmill
+
+Open-source platform for internal tools, workflows, API integrations, background jobs, and UIs. Rust backend + Svelte 5 frontend.
+
+## Workflow
+
+1. **Understand**: Before coding, explore the codebase (see Code Navigation below). Use `outline` to understand file structure, `body` to read specific symbols, `def`/`callers`/`callees` to trace code, `Grep` to find usages. Read `docs/` for domain context.
+2. **Plan**: For non-trivial changes, use plan mode. For large features, break into reviewable stages
+3. **Execute**: Follow coding patterns from skills (`rust-backend`, `svelte-frontend`)
+4. **Validate**: After every change, run the appropriate checks per `docs/validation.md`
+
+## Documentation
+
+- **Validation**: `docs/validation.md` — what checks to run based on what you changed
+- **Enterprise**: `docs/enterprise.md` — EE file conventions and PR workflow
+- **Backend patterns**: use the `rust-backend` skill when writing Rust code
+- **Frontend patterns**: use the `svelte-frontend` skill when writing Svelte code. Do NOT edit svelte files unless you have read that skill.
+- **Code review**: use `/local-review` to review a PR for bugs and CLAUDE.md compliance
+- **Domain guides**: `.claude/skills/native-trigger/` and `frontend/tutorial-system-guide.mdc`
+- **Brand/UI guidelines**: `frontend/brand-guidelines.md`
+
+## Dev Environment
+
+- **Backend**: `cargo run` from `backend/` (API at http://localhost:8000)
+- **Frontend**: `REMOTE=http://localhost:8000 npm run dev` from `frontend/` (port 3000+)
+- **DB**: `psql postgres://postgres:changeme@localhost:5432/windmill`
+- **Login**: `admin@windmill.dev` / `changeme`
+- **Instance settings**: navigate to `/#superadmin-settings`
+- **Migrations**: use `cargo sqlx migrate add -r <name>` from `backend/` to create new migrations (never generate timestamps manually)
+
+## Banned Patterns
+
+### `$bindable(default_value)` on optional props
+
+Using `$bindable(default_value)` on props that can be `undefined` is **banned**. This pattern causes subtle bugs because the default value masks the `undefined` state.
+
+**Bad:**
+
+```svelte
+let { my_prop = $bindable(default_value) }: { my_prop?: string } = $props()
+```
+
+**Correct alternatives:**
+
+1. **Use `$derived` with nullish coalescing** — handle the potential `undefined` at the usage site:
+
+   ```svelte
+   let { my_prop = $bindable() }: { my_prop?: string } = $props()
+   let effective_value = $derived(my_prop ?? default_value)
+   ```
+
+2. **Create a `useMyPropState()` helper** — encapsulate the undefined-handling logic in a reusable function and call it higher in the component tree, so the child component always receives a defined value.
+
+## Code Navigation
+
+`wm-ts-nav` is an AST-aware code navigator. Use **wm-ts-nav** for structural queries — it skips comments/strings and understands symbol boundaries.
+
+**MUST use `outline` before `Read`** on unfamiliar files — a 500-line file costs ~500 lines of context, while `outline` costs ~20. Then **MUST use `body "X"`** instead of reading a full file to see one function/struct. Use `Read` with offset/limit only when you need surrounding context that `body` doesn't capture.
+- `refs "X" --caller` instead of reading files to find which function contains each reference
+- `callers "X"` / `callees "X"` for call-graph questions
+
+EE files (`*_ee.rs`, `*_ee.ts`, `*_ee.svelte`) are indexed — you can `outline`, `def`, `body`, `refs` etc. on them just like regular files.
+
+```bash
+NAV="sh wm-ts-nav/nav"
+# Use --root backend for Rust, --root frontend/src for TS/Svelte
+$NAV --root backend outline backend/path/to/file.rs      # file structure
+$NAV --root backend def "ServiceName"                     # find definition
+$NAV --root backend body "decrypt_oauth_data"             # extract source code
+$NAV --root backend search "%" --parent ServiceName       # methods on a type
+$NAV --root backend search "Trigger" --kind struct        # find by kind
+$NAV --root backend refs "X" --file handler.rs --caller   # scoped refs with caller
+$NAV --root backend callers "X"                           # who calls X?
+$NAV --root backend callees "X"                           # what does X call?
+```
+
+**Limitations** — syntax-level analysis, no type inference. Use **Grep** instead when completeness matters (finding all usages, exhaustiveness checks):
+- `refs`/`callers`/`callees` can't follow re-exports, glob imports, or different import paths to the same symbol
+- Trait impls, macro-generated symbols (`sqlx::FromRow`), and namespace member access (`ns.X`) are invisible
+- `callees` shows all identifiers in a function body, not just actual calls
+
+## Core Principles
+
+- **MUST `outline` before `Read`** on unfamiliar files — then `body` or `Read` with offset/limit for specifics
+- Search for existing code to reuse before writing new code
+- Follow established patterns in the codebase
+- Keep changes focused — don't refactor beyond what's asked

CONTEXT.md

@@ -1,38 +0,0 @@
-# Windmill
-
-Open-source platform for internal tools, workflows, API integrations, background jobs and UIs. This file pins the vocabulary that is specific to Windmill's domain, so that code, docs and reviews name the same thing the same way.
-
-## Language
-
-### Flows
-
-**Step**:
-One node of a flow — the unit a user selects in the graph and configures in the right-hand panel. Typed as `FlowModule` in code.
-_Avoid_: module (ambiguous with the architectural sense), node, action
-
-**Step setting**:
-A per-step runtime option stored on the step itself: retries, error handling, timeout, concurrency limit, priority, cache, debounce, early stop, skip, suspend, sleep, lifetime. Distinct from the step's inputs and its code. The panel that edits them is the **run settings** tab; a single setting is still a step setting.
-_Avoid_: advanced setting, step config, flow option
-
-**Configured**:
-Said of a step setting whose config object is present on the step. Deliberately not the same as "would change the runtime's behaviour" — a setting can be configured and still be a no-op (`sleep` of `0`). Every surface that answers "is this setting on?" answers it this way.
-_Avoid_: enabled, active, effective
-
-**Trigger step**:
-The first step of a polling flow. It runs on a schedule and returns the items found since its last run; an empty return means there is nothing to process and the flow stops early, marked skipped rather than failed.
-_Avoid_: poll script, trigger node, schedule step
-
-**Default predicate**:
-The `stop_after_if` expression seeded onto a trigger step at creation, encoding what "nothing new" looks like. One value, owned in one place, shared by every path that creates a trigger step.
-
-**Connect**:
-Arming an input so that the next property picked fills it. A property can be picked from the prop picker or, when the panel is docked beside the graph, by clicking a step node's output. At most one input is armed per panel, so a pick always has exactly one destination.
-_Avoid_: link, bind, plug (the icon is a plug; the action is connecting)
-
-**Step input**:
-One argument of a step, edited in the step's input form. Its prop picker is a pane beside the form, always visible, so previous results can be browsed without connecting.
-_Avoid_: argument field, param
-
-**Expression input**:
-Any other place a property can be picked into: the loop iterator, skip and early-stop predicates, the retry condition, a branch predicate, timeout. Its prop picker opens in a popover from the connect button rather than taking a pane.
-_Avoid_: JS field, code input

Caddyfile

@@ -1,27 +1,18 @@
 {
 	layer4 {
 		:25 {
-			route {
-				proxy {
-					upstream windmill_server:2525
-				}
+			proxy {
+				to windmill_server:2525
 			}
 		}
 	}
 }
 
 {$BASE_URL} {
-        # Default to all interfaces (IPv4 + IPv6) when ADDRESS is unset. A bare
-        # `bind {$ADDRESS}` with an empty value makes Caddy >= 2.9 drop this whole
-        # site, silently disabling the HTTP proxy while the :25 layer4 listener stays up.
-        bind {$ADDRESS:0.0.0.0 ::}
+        bind {$ADDRESS}
 
-        # Extra services: LSP, Multiplayer, Debugger (windmill_extra gateway).
-        # reverse_proxy only reads its first argument as a matcher, so listing
-        # several paths inline turns the rest into upstream addresses. The paths
-        # have to go through a named matcher.
-        @extra path /ws/* /ws_mp/* /ws_debug/*
-        reverse_proxy @extra http://windmill_extra:3000
+        # Extra services: LSP, Multiplayer, Debugger (windmill_extra gateway)
+        reverse_proxy /ws/* /ws_mp/* /ws_debug/* http://windmill_extra:3000
 
         # Search indexer, Enterprise Edition (windmill_indexer:8002)
         # reverse_proxy /api/srch/* http://windmill_indexer:8002

Dockerfile

@@ -1,7 +1,7 @@
-ARG DEBIAN_IMAGE=debian:trixie-slim
-ARG RUST_IMAGE=rust:1.97-slim-trixie
+ARG DEBIAN_IMAGE=debian:bookworm-slim
+ARG RUST_IMAGE=rust:1.93-slim-bookworm
 
-FROM debian:trixie-slim AS nsjail
+FROM debian:bookworm-slim AS nsjail
 
 WORKDIR /nsjail
 
@@ -9,12 +9,12 @@ RUN apt-get -y update \
     && apt-get install -y \
     bison=2:3.8.* \
     flex=2.6.* \
-    g++=4:14.2.* \
-    gcc=4:14.2.* \
-    git=1:2.47.* \
+    g++=4:12.2.* \
+    gcc=4:12.2.* \
+    git=1:2.39.* \
     libprotobuf-dev=3.21.* \
     libnl-route-3-dev=3.7.* \
-    make=4.4.* \
+    make=4.3-4.1 \
     pkg-config=1.8.* \
     protobuf-compiler=3.21.*
 
@@ -44,7 +44,7 @@ FROM rust_base AS windmill_duckdb_ffi_internal_builder
 
 WORKDIR /windmill-duckdb-ffi-internal
 
-RUN apt-get update && apt-get install -y clang=1:19.0* libclang-dev=1:19.0* cmake=3.31.* && \
+RUN apt-get update && apt-get install -y clang=1:14.0-55.* libclang-dev=1:14.0-55.* cmake=3.25.* && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
@@ -66,7 +66,6 @@ RUN npm ci
 COPY frontend .
 RUN mkdir /backend
 COPY /backend/windmill-api/openapi.yaml /backend/windmill-api/openapi.yaml
-COPY /backend/oauth_connect.json /backend/oauth_connect.json
 COPY /openflow.openapi.yaml /openflow.openapi.yaml
 COPY /backend/windmill-api/build_openapi.sh /backend/windmill-api/build_openapi.sh
 COPY /system_prompts/auto-generated /system_prompts/auto-generated
@@ -79,8 +78,6 @@ COPY /python-client/docs/ /frontend/static/pydocs/
 RUN npm run generate-backend-client
 ENV NODE_OPTIONS "--max-old-space-size=8192"
 ARG VITE_BASE_URL ""
-# Must be declared for the build-arg to reach the bundle. See frontend/svelte.config.js.
-ARG WM_BUILD_VERSION=""
 # Read more about macro in docker/dev.nu
 # -- MACRO-SPREAD-WASM-PARSER-DEV-ONLY -- #
 RUN npm run build
@@ -100,7 +97,7 @@ ARG features=""
 
 COPY --from=planner /windmill/recipe.json recipe.json
 
-RUN apt-get update && apt-get install -y libxml2-dev=2.12.* libxmlsec1-dev=1.2.* libkrb5-dev libsasl2-dev libcurl4-openssl-dev clang=1:19.0* libclang-dev=1:19.0* cmake=3.31.* && \
+RUN apt-get update && apt-get install -y libxml2-dev=2.9.* libxmlsec1-dev=1.2.* libkrb5-dev libsasl2-dev libcurl4-openssl-dev clang=1:14.0-55.* libclang-dev=1:14.0-55.* cmake=3.25.* && \
     apt-get clean && \
     rm -rf /var/lib/apt/lists/*
 
@@ -137,8 +134,9 @@ FROM ${DEBIAN_IMAGE}
 
 ARG TARGETPLATFORM
 ARG POWERSHELL_VERSION=7.5.0
-ARG KUBECTL_VERSION=1.36.2
-ARG HELM_VERSION=3.21.2
+ARG POWERSHELL_DEB_VERSION=7.5.0-1
+ARG KUBECTL_VERSION=1.28.7
+ARG HELM_VERSION=3.14.3
 # NOTE: If changing, also change go version in workspace dependencies template at WorkspaceDependenciesEditor.svelte
 ARG GO_VERSION=1.26.0
 ARG APP=/usr/src/app
@@ -164,15 +162,14 @@ ENV PATH /usr/local/bin:/root/.local/bin:/tmp/.local/bin:$PATH
 
 
 RUN apt-get update \
-    && apt-get upgrade -y \
-    && apt-get install -y --no-install-recommends netbase tzdata ca-certificates wget curl jq unzip build-essential unixodbc xmlsec1 tini gnupg libargon2-1 \
+    && apt-get install -y --no-install-recommends netbase tzdata ca-certificates wget curl jq unzip build-essential unixodbc xmlsec1 software-properties-common tini gnupg lsb-release \
     && if echo "$features" | grep -q "ee"; then apt-get install -y --no-install-recommends libsasl2-modules-gssapi-mit krb5-user; fi \
     && apt-get clean \
     && rm -rf /var/lib/apt/lists/*
 
 # Install latest PostgreSQL client (pg_dump) from official PostgreSQL apt repository
 RUN curl -fsSL https://www.postgresql.org/media/keys/ACCC4CF8.asc | gpg --dearmor -o /usr/share/keyrings/postgresql-archive-keyring.gpg \
-    && echo "deb [signed-by=/usr/share/keyrings/postgresql-archive-keyring.gpg] https://apt.postgresql.org/pub/repos/apt $(. /etc/os-release; echo "$VERSION_CODENAME")-pgdg main" > /etc/apt/sources.list.d/pgdg.list \
+    && echo "deb [signed-by=/usr/share/keyrings/postgresql-archive-keyring.gpg] https://apt.postgresql.org/pub/repos/apt $(lsb_release -cs)-pgdg main" > /etc/apt/sources.list.d/pgdg.list \
     && apt-get update \
     && apt-get install -y --no-install-recommends postgresql-client \
     && apt-get clean \
@@ -185,14 +182,12 @@ RUN if [ "$WITH_GIT" = "true" ]; then \
     && rm -rf /var/lib/apt/lists/*; \
     else echo 'Building the image without git'; fi;
 
-# PowerShell ships as a tarball: the upstream .deb depends on libicu<=74 which no longer exists in trixie
 RUN if [ "$WITH_POWERSHELL" = "true" ]; then \
-    case "$TARGETPLATFORM" in \
-    "linux/amd64") pwsh_arch=x64 ;; \
-    "linux/arm64") pwsh_arch=arm64 ;; \
-    *) pwsh_arch="" ;; \
-    esac; \
-    if [ -n "$pwsh_arch" ]; then apt-get update -y && apt install libicu76 -y && wget -O powershell.tar.gz "https://github.com/PowerShell/PowerShell/releases/download/v${POWERSHELL_VERSION}/powershell-${POWERSHELL_VERSION}-linux-${pwsh_arch}.tar.gz" && apt-get clean \
+    if [ "$TARGETPLATFORM" = "linux/amd64" ]; then apt-get update -y && apt install libicu-dev -y && wget -O 'pwsh.deb' "https://github.com/PowerShell/PowerShell/releases/download/v${POWERSHELL_VERSION}/powershell_${POWERSHELL_DEB_VERSION}.deb_amd64.deb" && apt-get clean \
+    && rm -rf /var/lib/apt/lists/* && \
+    dpkg --install 'pwsh.deb' && \
+    rm 'pwsh.deb'; \
+    elif [ "$TARGETPLATFORM" = "linux/arm64" ]; then apt-get update -y && apt install libicu-dev -y && wget -O powershell.tar.gz "https://github.com/PowerShell/PowerShell/releases/download/v${POWERSHELL_VERSION}/powershell-${POWERSHELL_VERSION}-linux-arm64.tar.gz" && apt-get clean \
     && rm -rf /var/lib/apt/lists/* && \
     mkdir -p /opt/microsoft/powershell/7 && \
     tar zxf powershell.tar.gz -C /opt/microsoft/powershell/7 && \
@@ -237,14 +232,11 @@ ENV PATH="${PATH}:/usr/local/go/bin"
 ENV GO_PATH=/usr/local/go/bin/go
 
 # Install UV
-RUN curl --proto '=https' --tlsv1.2 -LsSf https://github.com/astral-sh/uv/releases/download/0.11.24/uv-installer.sh | sh && mv /root/.local/bin/uv /usr/local/bin/uv
+RUN curl --proto '=https' --tlsv1.2 -LsSf https://github.com/astral-sh/uv/releases/download/0.9.24/uv-installer.sh | sh && mv /root/.local/bin/uv /usr/local/bin/uv
 
 # Preinstall python runtimes to temp build location (will copy with world-writable perms later)
-# --compile-bytecode precompiles the stdlib to .pyc so jobs don't recompile it on every run
-# under the read-only nsjail runtime mount (uv >= 0.9.25). The copy below MUST preserve
-# timestamps or Python's mtime-based .pyc invalidation discards these compiled files.
-RUN UV_CACHE_DIR=/tmp/build_cache/uv UV_PYTHON_INSTALL_DIR=/tmp/build_cache/py_runtime uv python install 3.11 --compile-bytecode
-RUN UV_CACHE_DIR=/tmp/build_cache/uv UV_PYTHON_INSTALL_DIR=/tmp/build_cache/py_runtime uv python install $LATEST_STABLE_PY --compile-bytecode
+RUN UV_CACHE_DIR=/tmp/build_cache/uv UV_PYTHON_INSTALL_DIR=/tmp/build_cache/py_runtime uv python install 3.11
+RUN UV_CACHE_DIR=/tmp/build_cache/uv UV_PYTHON_INSTALL_DIR=/tmp/build_cache/py_runtime uv python install $LATEST_STABLE_PY
 
 
 RUN curl -sL https://deb.nodesource.com/setup_20.x | bash -
@@ -266,7 +258,7 @@ RUN export GOCACHE=/tmp/build_cache/go && \
 # chmod a+rw adds read+write WITHOUT removing execute bits (755->777, 644->666)
 # Note: uv python install only creates py_runtime, not uv cache - we create uv/go dirs for runtime
 RUN mkdir -p /tmp/windmill/cache && \
-    cp -r --preserve=timestamps /tmp/build_cache/* /tmp/windmill/cache/ && \
+    cp -r /tmp/build_cache/* /tmp/windmill/cache/ && \
     chmod -R a+rw /tmp/windmill/cache && \
     rm -rf /tmp/build_cache && \
     mkdir -p -m 777 /tmp/windmill/cache/uv /tmp/windmill/cache/go /tmp/windmill/cache/rustup /tmp/windmill/cache/cargo
@@ -296,7 +288,7 @@ RUN bun install -g windmill-cli \
 RUN curl -fsSL https://claude.ai/install.sh | bash \
     && cp /root/.local/share/claude/versions/* /usr/bin/claude
 
-COPY --from=php:8.3.30-cli-trixie /usr/local/bin/php /usr/bin/php
+COPY --from=php:8.3.30-cli-bookworm /usr/local/bin/php /usr/bin/php
 COPY --from=composer:2.9.5 /usr/bin/composer /usr/bin/composer
 
 # add the docker client to call docker from a worker if enabled
@@ -307,20 +299,10 @@ ENV CARGO_HOME="/tmp/windmill/cache/cargo"
 ENV LD_LIBRARY_PATH="."
 
 # nsjail runtime deps and binary
-RUN apt-get update && apt-get install -y --no-install-recommends libprotobuf32t64 libnl-route-3-200 libnl-3-200 \
+RUN apt-get update && apt-get install -y libprotobuf-dev libnl-route-3-dev \
     && apt-get clean && rm -rf /var/lib/apt/lists/*
 COPY --from=nsjail /nsjail/nsjail /bin/nsjail
 
-# crane: pulls + flattens images for the sandboxed container runtime (`# sandbox <image>`).
-# Single static binary — no daemon/store/root needed. See docs/docker-v2-runtime.md.
-ARG CRANE_VERSION=v0.21.7
-RUN arch="$(dpkg --print-architecture)"; \
-    case "$arch" in amd64) crane_arch=x86_64 ;; arm64) crane_arch=arm64 ;; *) echo >&2 "error: unsupported arch '$arch' for crane"; exit 1 ;; esac; \
-    wget -O /tmp/crane.tgz "https://github.com/google/go-containerregistry/releases/download/${CRANE_VERSION}/go-containerregistry_Linux_${crane_arch}.tar.gz" \
-    && tar -xzf /tmp/crane.tgz -C /usr/local/bin crane \
-    && rm /tmp/crane.tgz \
-    && chmod +x /usr/local/bin/crane
-
 WORKDIR ${APP}
 
 RUN ln -s ${APP}/windmill /usr/local/bin/windmill

REVIEW.md

@@ -1,69 +0,0 @@
-# Pull request review — shared policy
-
-You are reviewing a GitHub pull request for this repository. Apply this policy alongside your tool's output requirements.
-
-## Read the project rules first
-
-- Read `AGENTS.md` (repo root) and any `AGENTS.md` in directories touched by the diff before reviewing — they are the canonical contributor guide.
-- Quote the exact rule from `AGENTS.md` when flagging a violation.
-
-## Verdict (first line of the review)
-
-Start every review with a single verdict line, before any other section (the only thing that may appear above the verdict is the optional `cc @<PR_AUTHOR>` ping described in "Pinging the author" below). Pick exactly one:
-
-- **Good to merge** — no blocking issues and no nits worth surfacing.
-- **Mergeable, but should ideally address nits: <short list>** — no blockers, but P2 findings that are worth a look. The list must name each nit briefly (e.g. "doc/code mismatch in `foo.rs`, half-finished `pub fn bar`").
-- **Should address issues before merging: <short list>** — at least one P0 or P1 finding. The list must name each blocking issue briefly (e.g. "missing auth check on new `/api/x` handler, SQL injection in `build_query`").
-
-The names in the list must match findings detailed later in the review. If you list a nit or issue here, it must appear with full context in the body. Do not invent items that aren't in the body, and do not bury blockers in the body without surfacing them in the verdict.
-
-## Pinging the author
-
-If the prompt context provides a `PR AUTHOR` (GitHub login) and the verdict is NOT "Good to merge" (i.e. it is "Mergeable, but should ideally address nits: ..." or "Should address issues before merging: ..."), prepend a single line `cc @<PR_AUTHOR>` to the top-level review comment, above the verdict line. This pings the author so they get a notification that there are items to address. Skip the ping entirely when the verdict is "Good to merge" — there is nothing for the author to act on. Do not add the ping to inline comments; the top-level summary comment is the only place it belongs.
-
-## Review policy
-
-- Only report issues you are confident are real and introduced by this pull request.
-- Focus on bugs, security problems, performance, and clear `AGENTS.md` violations.
-- Do not report style nits, speculative concerns, pre-existing issues, or anything a normal linter / typechecker would obviously catch.
-- Self-validate each finding before posting: "is this definitely a real issue?" If uncertain, discard it.
-- Read additional files only when the diff is not enough to validate a finding.
-- Do not modify any files.
-
-## Severity triage
-
-Tag each finding with a severity. Always report P0 and P1. Report P2 only when the diff invites it (a new `pub fn`, a new module, a new exported component, a meaningful refactor).
-
-- **P0** — RCE, auth bypass, data loss, secrets in code, SQL injection, path traversal, broken auth on a public surface.
-- **P1** — significant bug, missing auth/authorization check on a new public surface, blocking I/O on a likely async path, race condition, missing input validation on caller-controlled parameters, observable performance regression.
-- **P2** — wrong module placement, doc/code mismatch, half-finished public abstractions (`pub fn` + `#[allow(dead_code)]` + `TODO`), `AGENTS.md` style violations, naming that contradicts the function's behavior.
-
-## Checklist for new public surfaces
-
-For any new `pub fn` / `pub async fn` / exported Svelte component / exported prop introduced by this PR, verify:
-
-- (a) auth/authorization expectations are documented in the doc comment OR enforced in the function body. A new `pub fn` that touches workspace data, secrets, files, or processes without an auth check or documented "caller MUST verify" contract is a P1.
-- (b) the function is placed in a module whose stated purpose matches what it does. Check the module-level doc comment (`//!`) — a config-file reader inside `external_ip.rs` is a P2.
-- (c) it is not half-finished. `pub fn` + `#[allow(dead_code)]` + a `TODO` is a smell that says the function should land together with its caller, not separately. Cite the relevant `AGENTS.md` rule.
-- (d) input validation defends against injection / traversal / overflow / NUL bytes at every parameter that may be caller-controlled.
-
-## Test coverage assessment
-
-End your review with a short "Test coverage" section calibrated to the layers actually changed by the diff. Skip categories the PR does not touch.
-
-- **Backend** (Rust under `backend/`) — expect Rust unit tests for new logic. For new or modified API handlers, worker steps, queue/cron behavior, or DB access, also expect or note the absence of integration tests. Pure-refactor backend PRs don't need new tests if existing tests cover the surface.
-- **Frontend** (Svelte / TS under `frontend/`) — the codebase does not generally test Svelte components, so do not ask for component tests. Only flag missing tests for new pure-logic utilities (the kind of file that already has a sibling `*.test.ts`, e.g. `flowDiff`, `previousResults`, copilot logic).
-- **CI / workflows / docs / config-only** — no automated tests expected; say so explicitly so the reader knows you considered it.
-
-Then state what manual verification, if any, is still needed before merge:
-
-- Describe each manual scenario as a short paragraph (not a numbered list): what page / action / input, and what observable outcome confirms correctness.
-- If the diff has no in-app surface to exercise (purely backend internals, CI, docs, or refactor), say that plainly.
-
-## Additional reviewer instructions
-
-If the prompt or context includes an "Additional reviewer instructions" section, treat it as extra guidance from the human who triggered this review and follow it.
-
-## Prior PR discussion
-
-If the prompt or context includes a "Prior PR discussion" section, this PR has already received review activity. Look for your own previous comment, take it into account, focus on what changed in the latest commits, and do not repeat findings the human already pushed back on or addressed.

ai_evals/.gitignore

@@ -1,2 +0,0 @@
-.env
-results/

ai_evals/AGENTS.md

@@ -1,14 +0,0 @@
-# AI Evals
-
-Black-box benchmark cases for the Windmill AI generation modes (`flow`, `app`,
-`script`, `cli`, `global`).
-
-**Authoring and running cases is documented in the `ai-evals` skill** — load it
-before adding/changing a case or running a benchmark. Claude Code reads
-`.claude/skills/ai-evals/SKILL.md`; Codex and Pi read
-`.agents/skills/ai-evals/SKILL.md` (same canonical file). Invoke with `/ai-evals` in
-Claude Code, `$ai-evals` in Codex, or `pi --skill ai-evals`.
-
-For AI chat / copilot changes that these evals measure, see the `ai-chat` skill.
-
-The full case format, fields, and fixture details remain in `ai_evals/README.md`.

ai_evals/CLAUDE.md

@@ -1 +0,0 @@
-@AGENTS.md

ai_evals/README.md

@@ -1,260 +0,0 @@
-# AI Evals
-
-Small benchmark runner for the Windmill AI generation modes:
-
-- `cli`
-- `flow`
-- `script`
-- `app`
-- `global`
-
-The benchmark always tests the current production prompts, tools, and guidance in this checkout.
-
-Each attempt runs:
-
-1. the real production path
-2. deterministic validation
-3. LLM judging
-
-## Install
-
-```bash
-cd ai_evals
-bun install
-```
-
-Frontend modes also require frontend dependencies:
-
-```bash
-cd frontend
-bun install
-```
-
-## Commands
-
-List model aliases:
-
-```bash
-cd ai_evals
-bun run cli -- models
-```
-
-List cases:
-
-```bash
-cd ai_evals
-bun run cli -- cases
-bun run cli -- cases flow
-```
-
-Run benchmarks:
-
-```bash
-cd ai_evals
-bun run cli -- run flow
-bun run cli -- run flow flow-test4-order-processing-loop --model opus
-bun run cli -- run flow flow-test0-sum-two-numbers --models haiku,opus,4o
-bun run cli -- run flow flow-test0-sum-two-numbers --runs 3 --verbose
-bun run cli -- run flow --record
-GEMINI_API_KEY=... bun run cli -- run app app-test1-counter-create --model gemini-3-flash-preview
-WMILL_AI_EVAL_BACKEND_URL=http://127.0.0.1:8000 bun run cli -- run flow --backend-validation preview
-bun run cli -- run global global-test1-script-create
-bun run cli -- run cli bun-hello-script
-```
-
-Public CLI surface:
-
-- `models`
-- `cases [mode]`
-- `run <mode> [caseIds...]`
-
-`run` options:
-
-- `--runs <n>`: repeat each case `n` times
-- `--output <path>`: custom result JSON path
-- `--model <alias>`: choose the model under test
-- `--models <a,b,c>`: run the same cases sequentially against several model aliases
-- `--verbose`: stream assistant output for frontend runs
-- `--skip-judge`: skip LLM judge scoring for the run
-- `--execution-only`: only require the model/proxy/frontend loop to complete; skip validators, tool expectations, backend artifact validation, and judge scoring
-- `--record`: append a compact tracked summary line to `ai_evals/history/<mode>.jsonl` for full-suite runs only
-- `--backend-validation <mode>`: optional backend smoke validation (`off` or `preview`) for `script` and `flow` evals
-
-## Models
-
-Use `bun run cli -- models` to see the current aliases.
-
-Today:
-
-- `haiku`
-- `sonnet`
-- `opus`
-- `4o`
-- `gpt-5.5`
-- `gemini-3-flash-preview`
-- `gemini-3.1-pro-preview`
-- `deepseek-v4-flash`
-- `deepseek-v4-pro`
-
-Notes:
-
-- the command also prints accepted alias spellings such as `gpt-4o`, `gpt-55`, `claude-opus-4.6`, and `claude-haiku-4.5`
-- frontend modes (`flow`, `script`, `app`, `global`) can use Anthropic, OpenAI, Gemini, and DeepSeek-backed aliases
-- `cli` mode always uses the Anthropic agent SDK, so only Anthropic aliases are valid there
-- the judge model is separate and currently defaults to `claude-sonnet-4-6`; use `--skip-judge` for deterministic-only runs
-
-## Case Format
-
-Cases live in one YAML file per mode under `ai_evals/cases/`.
-
-Minimal shape:
-
-```yaml
-- id: flow-test0-sum-two-numbers
-  prompt: |-
-    Create a flow that takes two numbers, `a` and `b`, and returns their sum.
-  initial: ai_evals/fixtures/...
-  expected: ai_evals/fixtures/...
-```
-
-Optional fields:
-
-- `initial`: starting state fixture
-- `expected`: expected artifact fixture
-- `validate`: extra deterministic validation rules
-- `runtime.backendPreview`: optional real backend preview config for smoke validation
-
-For `flow` mode, `validate` can express requirements such as:
-
-- accepted input schema shapes
-- required `results.*` reference validity
-- required module/code/input characteristics
-
-For `app` mode, `validate` can express narrow hard requirements such as:
-
-- required frontend file paths or backend runnable keys
-- minimum backend runnable counts
-- required backend runnable types
-- minimum datatable / datatable-table counts
-- specific required datatable tables
-
-For `global` mode, `validate` can express draft-level requirements such as:
-
-- required draft type/path/language
-- required or forbidden snippets in draft values
-- required or forbidden draft counts
-- forbidden draft paths
-
-Global initial fixtures can also seed `liveEditorDrafts` with `type`,
-`storagePath`, `effectivePath`, and `value` fields. These drafts emulate the
-currently open script, flow, or raw app editor so cases can test prompts that
-refer to "this" or the "current" item.
-
-Global (and flow) initial fixtures can seed `workspace.datatables` so the
-`list_datatables`, `get_datatable_table_schema`, and `exec_datatable_sql` tools
-return seeded data during evals. Each entry is
-`{ datatable_name, schemas: { <schema>: { <table>: { columns, rows? } } } }`.
-SQL runs through a small in-memory engine (`datatableSqlEngine.ts`), not a real
-database. Writes are **stateful within a case**: `CREATE`/`DROP`/`INSERT`/`UPDATE`/
-`DELETE` mutate the seeded datatable in place, so a later `list_datatables`,
-`get_datatable_table_schema`, `SELECT`, or `information_schema` query reflects them
-— this is what stops a model from looping when it re-queries to verify a write.
-The engine is best-effort: `SELECT` returns all rows of the referenced (or first)
-table with no WHERE filtering/projection/joins, `WHERE` on UPDATE/DELETE supports
-`col = value` predicates joined by `AND`, and anything unparseable is a no-op
-success. So validate datatable cases through tool-use and SQL-argument assertions
-(`requiredToolsUsed`, `stringIncludesAnyOf`) — not through exact returned row
-values. An empty/absent `datatables` seed makes `list_datatables` return `[]`,
-which is what the "no datatable configured" blocking cases rely on.
-
-Set `WMILL_AI_EVAL_DISABLE_ACTIVE_EDITOR_CONTEXT=1` to run those cases with
-the old behavior where the live editor is only discoverable through
-`list_workspace_items`.
-
-App fixtures can also include an optional `datatables.json` file at the fixture root.
-
-For `flow` mode, an `initial` fixture can also include a benchmark workspace catalog of
-existing scripts and flows. That lets the real `search_workspace` and
-`get_runnable_details` tools discover reusable workspace runnables during evals.
-
-If `--backend-validation preview` is enabled:
-
-- `script` evals run a real backend script preview in an isolated temp workspace
-- `flow` evals run a real backend flow preview only for cases that define `runtime.backendPreview`
-- `flow` cases with `initial.workspace` fixtures seed those scripts and flows into the preview workspace before preview
-- when `WMILL_AI_EVAL_BACKEND_WORKSPACE` is set, `ai_evals` creates or reuses that workspace as a dedicated test workspace, clears managed eval assets under `f/evals/*` before each preview run, and then reseeds the current case fixtures
-
-Supported backend env vars:
-
-- `WMILL_AI_EVAL_BACKEND_VALIDATION=preview`
-- `WMILL_AI_EVAL_BACKEND_URL=http://127.0.0.1:8000`
-- `WMILL_AI_EVAL_BACKEND_EMAIL=admin@windmill.dev`
-- `WMILL_AI_EVAL_BACKEND_PASSWORD=changeme`
-- `WMILL_AI_EVAL_BACKEND_WORKSPACE=integration-tests` to reuse an existing workspace on CE installs with low workspace limits
-
-Frontend modes require a reachable Windmill backend and send model requests through the workspace AI proxy at `/api/w/{workspace}/ai/proxy`. At startup, `ai_evals` checks the resolved backend URL and fails early with setup guidance if the backend cannot be reached or login fails.
-
-For frontend modes:
-
-- `ai_evals` creates a temporary backend workspace, or creates/reuses `WMILL_AI_EVAL_BACKEND_WORKSPACE` when it is set
-- it upserts a provider resource under `f/evals/ai/<provider>`
-- frontend requests go through `/api/w/{workspace}/ai/proxy`
-
-## Results And Artifacts
-
-Every run writes:
-
-- a summary JSON under `ai_evals/results/`
-- generated artifacts in a sibling directory
-
-If `--record` is used, the CLI also appends one compact JSON line to:
-
-- `ai_evals/history/flow.jsonl`
-- `ai_evals/history/script.jsonl`
-- `ai_evals/history/app.jsonl`
-- `ai_evals/history/global.jsonl`
-- `ai_evals/history/cli.jsonl`
-
-Each recorded line contains:
-
-- run metadata (`createdAt`, `gitSha`, `mode`, `runModel`, `judgeModel`)
-- suite totals (`caseCount`, `attemptCount`, `passedAttempts`, `passRate`, `averageDurationMs`, `averagePassedDurationMs`, `averageJudgeScore`)
-- average token usage (`averageTokenUsagePerAttempt`, `averageTokenUsagePerPassedAttempt`)
-- per-case metrics under `cases[]` (`averageDurationMs`, `averagePassedDurationMs`, `averageJudgeScore`, `averageTokenUsagePerAttempt`, `averageTokenUsagePerPassedAttempt`, pass rate)
-- `failedCaseIds`
-
-The CLI headline duration and token averages use passed attempts only.
-All-attempt averages are still recorded to make failures auditable without
-letting failed attempts skew success cost comparisons.
-
-Example:
-
-- summary: `ai_evals/results/2026-04-09T09-40-33.051Z__flow.json`
-- artifacts: `ai_evals/results/2026-04-09T09-40-33.051Z__flow/`
-
-Typical artifacts by mode:
-
-- `flow`: `flow.json`
-- `script`: `script.json` plus the generated script file
-- `app`: `app.json` plus frontend/backend files
-- `global`: `global-drafts.json`
-- `cli`: `assistant-output.txt`, `trace.json`, `wmill-invocations.jsonl`, plus generated workspace files
-- backend-validated attempts also include `backend-preview.json`
-
-## Layout
-
-- `cases/`: one YAML file per mode
-- `fixtures/`: initial and expected fixtures
-- `core/`: shared loading, model resolution, validation, judging, and result writing
-- `modes/`: one runner per mode
-- `history/`: optional tracked pass-rate history written by `run --record`, one JSONL file per mode
-- `results/`: local benchmark output and artifacts
-
-## Notes
-
-- Frontend modes reuse the production frontend chat code through the Vitest bridge.
-- Global mode evaluates the production global AI tools and validates the resulting AI draft store.
-- CLI mode creates an isolated workspace, writes the current checkout guidance into it, and benchmarks the real skills / `AGENTS.md` flow.
-- CLI mode now also records a structured trace of invoked skills, tool calls, proposed `wmill` commands, and any attempted `wmill` executions.
-- Frontend progress streams live while the benchmark is running.
-- Deterministic validators should stay focused on real correctness constraints, not one exact implementation shape.

ai_evals/adapters/cli/runtime.test.ts

@@ -1,149 +0,0 @@
-import { describe, expect, it } from "bun:test";
-import {
-  anthropicUsageToBenchmarkTokenUsage,
-  extractCliResultTokenUsage,
-  extractProposedWmillCommands,
-  parseWmillInvocationLog,
-} from "./runtime";
-
-describe("anthropicUsageToBenchmarkTokenUsage", () => {
-  it("includes cache tokens in prompt usage", () => {
-    expect(
-      anthropicUsageToBenchmarkTokenUsage({
-        input_tokens: 120,
-        output_tokens: 45,
-        cache_creation_input_tokens: 30,
-        cache_read_input_tokens: 5,
-      })
-    ).toEqual({
-      prompt: 155,
-      completion: 45,
-      total: 200,
-    });
-  });
-
-  it("returns null when usage is absent", () => {
-    expect(anthropicUsageToBenchmarkTokenUsage(null)).toBeNull();
-  });
-});
-
-describe("extractCliResultTokenUsage", () => {
-  it("reads aggregate usage from the SDK result event", () => {
-    expect(
-      extractCliResultTokenUsage({
-        type: "result",
-        usage: {
-          input_tokens: 400,
-          output_tokens: 120,
-          cache_creation_input_tokens: 50,
-          cache_read_input_tokens: 25,
-        },
-      })
-    ).toEqual({
-      prompt: 475,
-      completion: 120,
-      total: 595,
-    });
-  });
-
-  it("falls back to modelUsage when aggregate usage is unavailable", () => {
-    expect(
-      extractCliResultTokenUsage({
-        type: "result",
-        modelUsage: {
-          opus: {
-            inputTokens: 200,
-            outputTokens: 60,
-            cacheCreationInputTokens: 10,
-            cacheReadInputTokens: 5,
-          },
-          haiku: {
-            inputTokens: 80,
-            outputTokens: 20,
-            cacheCreationInputTokens: 0,
-            cacheReadInputTokens: 15,
-          },
-        },
-      })
-    ).toEqual({
-      prompt: 310,
-      completion: 80,
-      total: 390,
-    });
-  });
-});
-
-describe("extractProposedWmillCommands", () => {
-  it("extracts proposed commands from bullets, code blocks, and inline code", () => {
-    expect(
-      extractProposedWmillCommands(`
-Next:
-- \`wmill generate-metadata --yes\`
-- wmill sync push
-
-You can inspect failures with \`wmill job logs 123\`.
-`)
-    ).toEqual([
-      "wmill generate-metadata --yes",
-      "wmill sync push",
-      "wmill job logs 123",
-    ]);
-  });
-
-  it("extracts inline prose commands that are not wrapped in backticks", () => {
-    expect(
-      extractProposedWmillCommands(
-        "The first command is wmill sync pull before you edit locally."
-      )
-    ).toEqual(["wmill sync pull"]);
-  });
-
-  it("extracts multiple inline prose commands from a single sentence", () => {
-    expect(
-      extractProposedWmillCommands(
-        "Run wmill generate-metadata and then wmill sync push when you are ready."
-      )
-    ).toEqual(["wmill generate-metadata", "wmill sync push"]);
-  });
-
-  it("ignores negated command mentions", () => {
-    expect(
-      extractProposedWmillCommands(
-        "Do not run `wmill sync push`. Instead run `wmill sync pull` first."
-      )
-    ).toEqual(["wmill sync pull"]);
-  });
-});
-
-describe("parseWmillInvocationLog", () => {
-  it("parses stubbed wmill invocations into structured records", () => {
-    expect(
-      parseWmillInvocationLog(`noise
-__WMILL_BENCHMARK__
-2026-04-21T12:00:00+00:00
-/tmp/workspace
-2
-generate-metadata
---yes
-__WMILL_BENCHMARK__
-2026-04-21T12:00:05+00:00
-/tmp/workspace
-3
-sync
-push
---dry-run
-`)
-    ).toEqual([
-      {
-        argv: ["generate-metadata", "--yes"],
-        cwd: "/tmp/workspace",
-        timestamp: "2026-04-21T12:00:00+00:00",
-      },
-      {
-        argv: ["sync", "push", "--dry-run"],
-        cwd: "/tmp/workspace",
-        timestamp: "2026-04-21T12:00:05+00:00",
-      },
-    ]);
-  });
-});

ai_evals/adapters/cli/runtime.ts

@@ -1,543 +0,0 @@
-import { query, type Options } from "@anthropic-ai/claude-agent-sdk";
-import { chmod, mkdir, readFile, writeFile } from "node:fs/promises";
-import { delimiter, join } from "path";
-import { fileURLToPath } from "url";
-import { getCliEvalModel, resolveEvalModel, type CliEvalModelConfig } from "../../core/models";
-import type {
-  BenchmarkTokenUsage,
-  CliToolInvocation,
-  CliTrace,
-  CliWmillInvocation,
-} from "../../core/types";
-
-export type ToolInvocation = CliToolInvocation;
-
-export interface PromptRunResult {
-  output: string;
-  durationMs: number;
-  tokenUsage: BenchmarkTokenUsage | null;
-  // Input tokens on the last assistant turn. The SDK `result` message reports
-  // usage cumulatively, so the final context size comes from per-turn usage.
-  finalContextTokens: number | null;
-  trace: CliTrace;
-}
-
-interface AnthropicUsageLike {
-  input_tokens?: number | null;
-  output_tokens?: number | null;
-  cache_creation_input_tokens?: number | null;
-  cache_read_input_tokens?: number | null;
-}
-
-interface AnthropicModelUsageLike {
-  inputTokens?: number | null;
-  outputTokens?: number | null;
-  cacheCreationInputTokens?: number | null;
-  cacheReadInputTokens?: number | null;
-}
-
-interface CliResultMessageLike {
-  type?: string;
-  usage?: AnthropicUsageLike | null;
-  modelUsage?: Record<string, AnthropicModelUsageLike> | null;
-}
-
-const REPO_ROOT = fileURLToPath(new URL("../../../", import.meta.url));
-export const DEFAULT_CLI_EVAL_MODEL: CliEvalModelConfig = getCliEvalModel(resolveEvalModel("cli"));
-const WMILL_STUB_DIR_NAME = ".wmill-benchmark-bin";
-const WMILL_LOG_FILE_NAME = ".wmill-benchmark-wmill-invocations.log";
-const WMILL_LOG_MARKER = "__WMILL_BENCHMARK__";
-const NEGATED_COMMAND_PREFIX = /(?:^|\b)(?:do not|don't|dont|never|instead of)\s+(?:run|use)?\s*$/i;
-const COMMAND_STOP_WORDS = new Set([
-  "and",
-  "before",
-  "after",
-  "then",
-  "instead",
-  "otherwise",
-  "because",
-  "so",
-  "if",
-  "when",
-  "while",
-  "once",
-]);
-const COMMAND_STOP_TOKENS = new Set(["-", "–", "—", "|"]);
-
-export function getGeneratedSkillsSource(): string {
-  return join(REPO_ROOT, "system_prompts", "auto-generated", "skills");
-}
-
-export function anthropicUsageToBenchmarkTokenUsage(
-  usage: AnthropicUsageLike | null | undefined
-): BenchmarkTokenUsage | null {
-  if (!usage) {
-    return null;
-  }
-
-  const prompt =
-    (usage.input_tokens ?? 0) +
-    (usage.cache_creation_input_tokens ?? 0) +
-    (usage.cache_read_input_tokens ?? 0);
-  const completion = usage.output_tokens ?? 0;
-
-  return {
-    prompt,
-    completion,
-    total: prompt + completion,
-  };
-}
-
-export function extractCliResultTokenUsage(message: unknown): BenchmarkTokenUsage | null {
-  if (!message || typeof message !== "object") {
-    return null;
-  }
-
-  const resultMessage = message as CliResultMessageLike;
-  if (resultMessage.type !== "result") {
-    return null;
-  }
-
-  const usage = anthropicUsageToBenchmarkTokenUsage(resultMessage.usage);
-  if (usage) {
-    return usage;
-  }
-
-  if (!resultMessage.modelUsage || typeof resultMessage.modelUsage !== "object") {
-    return null;
-  }
-
-  let prompt = 0;
-  let completion = 0;
-  let sawModelUsage = false;
-
-  for (const modelUsage of Object.values(resultMessage.modelUsage)) {
-    if (!modelUsage || typeof modelUsage !== "object") {
-      continue;
-    }
-
-    prompt +=
-      (modelUsage.inputTokens ?? 0) +
-      (modelUsage.cacheCreationInputTokens ?? 0) +
-      (modelUsage.cacheReadInputTokens ?? 0);
-    completion += modelUsage.outputTokens ?? 0;
-    sawModelUsage = true;
-  }
-
-  if (!sawModelUsage) {
-    return null;
-  }
-
-  return {
-    prompt,
-    completion,
-    total: prompt + completion,
-  };
-}
-
-export async function runPromptAndCapture(
-  prompt: string,
-  cwd: string,
-  maxTurns: number = 3,
-  modelConfig: CliEvalModelConfig = DEFAULT_CLI_EVAL_MODEL
-): Promise<PromptRunResult> {
-  const toolsUsed: ToolInvocation[] = [];
-  const skillsInvoked: string[] = [];
-  const bashCommands: string[] = [];
-  let output = "";
-  let assistantMessageCount = 0;
-  let tokenUsage: BenchmarkTokenUsage | null = null;
-  let finalContextTokens: number | null = null;
-  const startedAt = Date.now();
-  const stubBinDir = join(cwd, WMILL_STUB_DIR_NAME);
-  const wmillLogPath = join(cwd, WMILL_LOG_FILE_NAME);
-
-  const options: Options = {
-    cwd,
-    model: modelConfig.model,
-    maxTurns,
-    settingSources: ["project"],
-    allowedTools: ["Skill", "Read", "Glob", "Grep", "Bash", "Write", "Edit"],
-    env: {
-      ...getQueryEnv(),
-      PATH: process.env.PATH ? `${stubBinDir}${delimiter}${process.env.PATH}` : stubBinDir,
-      WMILL_BENCHMARK_LOG_PATH: wmillLogPath,
-    },
-  };
-
-  await installWmillStub(stubBinDir);
-
-  for await (const message of query({ prompt, options })) {
-    if (message.type === "assistant") {
-      assistantMessageCount += 1;
-      const turnContext = anthropicUsageToBenchmarkTokenUsage(
-        message.message?.usage
-      )?.prompt;
-      if (turnContext && turnContext > 0) {
-        finalContextTokens = turnContext;
-      }
-      const content = message.message?.content;
-      if (Array.isArray(content)) {
-        for (const block of content) {
-          if (block.type === "tool_use") {
-            const input = normalizeToolInput(block.input);
-            toolsUsed.push({
-              tool: block.name,
-              input,
-              timestamp: Date.now()
-            });
-
-            if (block.name === "Skill") {
-              const skillInput = input as { skill?: string };
-              if (skillInput.skill) {
-                pushUnique(skillsInvoked, skillInput.skill);
-              }
-            }
-
-            if (block.name === "Bash") {
-              for (const command of extractBashCommands(input)) {
-                pushUnique(bashCommands, command);
-              }
-            }
-          } else if (block.type === "text") {
-            output += block.text;
-          }
-        }
-      }
-    } else if (message.type === "result") {
-      const resultMessage = message as { result?: string };
-      tokenUsage = extractCliResultTokenUsage(message) ?? tokenUsage;
-      if (typeof resultMessage.result === "string") {
-        output += resultMessage.result;
-      }
-    }
-  }
-
-  const proposedCommands = extractProposedWmillCommands(output);
-  const wmillInvocations = await readWmillInvocationLog(wmillLogPath);
-
-  return {
-    output,
-    durationMs: Date.now() - startedAt,
-    tokenUsage,
-    finalContextTokens,
-    trace: {
-      toolsUsed,
-      skillsInvoked,
-      assistantMessageCount,
-      bashCommands,
-      proposedCommands,
-      executedWmillCommands: wmillInvocations.map(formatExecutedWmillCommand),
-      wmillInvocations,
-      firstMutationToolIndex: getFirstMutationToolIndex(toolsUsed),
-    },
-  };
-}
-
-export function wasSkillInvoked(result: PromptRunResult, skillName: string): boolean {
-  return result.trace.skillsInvoked.some((skill) => skill === skillName);
-}
-
-export function wasToolUsed(result: PromptRunResult, toolName: string): boolean {
-  return result.trace.toolsUsed.some((tool) => tool.tool === toolName);
-}
-
-export function formatCliRunModelLabel(modelConfig: CliEvalModelConfig): string {
-  return `${modelConfig.provider}:${modelConfig.model}`;
-}
-
-export function getToolInputs(
-  result: PromptRunResult,
-  toolName: string
-): Record<string, unknown>[] {
-  return result.trace.toolsUsed
-    .filter((tool) => tool.tool === toolName)
-    .map((tool) => tool.input);
-}
-
-export function extractProposedWmillCommands(output: string): string[] {
-  const commands: string[] = [];
-
-  for (const line of output.split(/\r?\n/)) {
-    for (const command of extractInlineBacktickCommands(line)) {
-      pushUnique(commands, command);
-    }
-
-    for (const command of extractInlineProseCommands(line.replace(/^\s*(?:[-*]|\d+\.)\s*/, ""))) {
-      pushUnique(commands, command);
-    }
-  }
-
-  return commands;
-}
-
-export function parseWmillInvocationLog(raw: string): CliWmillInvocation[] {
-  const entries: CliWmillInvocation[] = [];
-  const lines = raw.split(/\r?\n/);
-
-  for (let index = 0; index < lines.length; index += 1) {
-    if (lines[index] !== WMILL_LOG_MARKER) {
-      continue;
-    }
-
-    const timestamp = lines[index + 1] ?? "";
-    const cwd = lines[index + 2] ?? "";
-    const argCount = Number.parseInt(lines[index + 3] ?? "", 10);
-    if (!Number.isFinite(argCount) || argCount < 0) {
-      continue;
-    }
-
-    const start = index + 4;
-    const argv = lines.slice(start, start + argCount);
-    entries.push({ argv, cwd, timestamp });
-    index = start + argCount - 1;
-  }
-
-  return entries;
-}
-
-async function installWmillStub(binDir: string): Promise<void> {
-  await mkdir(binDir, { recursive: true });
-
-  const stubPath = join(binDir, "wmill");
-  const script = `#!/usr/bin/env bash
-set -euo pipefail
-{
-  printf '${WMILL_LOG_MARKER}\\n'
-  date -u +"%Y-%m-%dT%H:%M:%SZ"
-  printf '%s\\n' "$PWD"
-  printf '%s\\n' "$#"
-  printf '%s\\n' "$@"
-} >> "\${WMILL_BENCHMARK_LOG_PATH:?}"
-printf 'wmill benchmark stub: do not execute Windmill CLI commands during ai_evals; describe them in the final response instead.\\n' >&2
-exit 97
-`;
-
-  await writeFile(stubPath, script, "utf8");
-  await chmod(stubPath, 0o755);
-}
-
-async function readWmillInvocationLog(logPath: string): Promise<CliWmillInvocation[]> {
-  const raw = await readFile(logPath, "utf8").catch(() => null);
-  if (!raw) {
-    return [];
-  }
-  return parseWmillInvocationLog(raw);
-}
-
-function getQueryEnv(): Record<string, string> {
-  return Object.fromEntries(
-    Object.entries(process.env).flatMap(([key, value]) =>
-      typeof value === "string" ? [[key, value]] : []
-    )
-  );
-}
-
-function normalizeToolInput(input: unknown): Record<string, unknown> {
-  if (input && typeof input === "object" && !Array.isArray(input)) {
-    return input as Record<string, unknown>;
-  }
-
-  if (typeof input === "string") {
-    return { raw: input };
-  }
-
-  return {};
-}
-
-function extractBashCommands(input: Record<string, unknown>): string[] {
-  const commands: string[] = [];
-
-  for (const key of ["command", "cmd", "script", "raw"]) {
-    const value = input[key];
-    if (typeof value === "string") {
-      for (const line of value.split(/\r?\n/)) {
-        const command = normalizeCommandCandidate(line);
-        if (command) {
-          pushUnique(commands, command);
-        }
-      }
-    }
-  }
-
-  return commands;
-}
-
-function extractInlineBacktickCommands(line: string): string[] {
-  const commands: string[] = [];
-  const regex = /`(wmill [^`\n]+)`/g;
-  let match: RegExpExecArray | null = null;
-
-  while ((match = regex.exec(line)) !== null) {
-    if (hasNegatedCommandPrefix(line.slice(0, match.index))) {
-      continue;
-    }
-
-    const command = normalizeCommandCandidate(match[1]);
-    if (command) {
-      pushUnique(commands, command);
-    }
-  }
-
-  return commands;
-}
-
-function extractInlineProseCommands(line: string): string[] {
-  const commands: string[] = [];
-  let searchFrom = 0;
-
-  while (true) {
-    const inlineIndex = line.toLowerCase().indexOf("wmill ", searchFrom);
-    if (inlineIndex === -1) {
-      return commands;
-    }
-
-    if (!hasNegatedCommandPrefix(line.slice(0, inlineIndex))) {
-      const command = extractInlineProseCommandAt(line, inlineIndex);
-      if (command) {
-        pushUnique(commands, command);
-      }
-    }
-
-    searchFrom = inlineIndex + "wmill ".length;
-  }
-}
-
-function extractInlineProseCommandAt(line: string, startIndex: number): string | null {
-  const tokens = ["wmill"];
-  let cursor = startIndex + "wmill".length;
-
-  while (cursor < line.length) {
-    while (cursor < line.length && /\s/.test(line[cursor]!)) {
-      cursor += 1;
-    }
-
-    if (cursor >= line.length) {
-      break;
-    }
-
-    const current = line[cursor]!;
-    if ("`.,;:()[]{}".includes(current)) {
-      break;
-    }
-
-    const token = readCommandToken(line, cursor);
-    if (!token) {
-      break;
-    }
-
-    if (COMMAND_STOP_WORDS.has(token.value.toLowerCase())) {
-      break;
-    }
-
-    if (COMMAND_STOP_TOKENS.has(token.value)) {
-      break;
-    }
-
-    tokens.push(token.value);
-    cursor = token.nextIndex;
-  }
-
-  if (tokens.length <= 1) {
-    return null;
-  }
-
-  return normalizeCommandCandidate(tokens.join(" "));
-}
-
-function readCommandToken(
-  line: string,
-  startIndex: number
-): { value: string; nextIndex: number } | null {
-  const firstChar = line[startIndex]!;
-
-  if (firstChar === `"` || firstChar === `'`) {
-    const endIndex = line.indexOf(firstChar, startIndex + 1);
-    const nextIndex = endIndex === -1 ? line.length : endIndex + 1;
-    return {
-      value: line.slice(startIndex, nextIndex),
-      nextIndex,
-    };
-  }
-
-  if (firstChar === "<") {
-    const endIndex = line.indexOf(">", startIndex + 1);
-    const nextIndex = endIndex === -1 ? line.length : endIndex + 1;
-    return {
-      value: line.slice(startIndex, nextIndex),
-      nextIndex,
-    };
-  }
-
-  let endIndex = startIndex;
-  while (endIndex < line.length && !/[\s`.,;:()[\]{}#]/.test(line[endIndex]!)) {
-    endIndex += 1;
-  }
-
-  if (endIndex === startIndex) {
-    return null;
-  }
-
-  return {
-    value: line.slice(startIndex, endIndex),
-    nextIndex: endIndex,
-  };
-}
-
-function hasNegatedCommandPrefix(prefix: string): boolean {
-  const normalizedPrefix = prefix
-    .toLowerCase()
-    .replace(/[`"'“”‘’]/g, " ")
-    .replace(/\s+/g, " ")
-    .trimEnd();
-
-  return NEGATED_COMMAND_PREFIX.test(normalizedPrefix);
-}
-
-function normalizeCommandCandidate(value: string): string | null {
-  const trimmed = value.trim().replace(/^`|`$/g, "");
-  if (!trimmed) {
-    return null;
-  }
-
-  const normalized = trimmed
-    .replace(/\s+/g, " ")
-    .replace(/[`.;:,]+$/g, "")
-    .trim();
-
-  return normalized.length > 0 ? normalized : null;
-}
-
-function formatExecutedWmillCommand(entry: CliWmillInvocation): string {
-  return ["wmill", ...entry.argv].join(" ").trim();
-}
-
-function getFirstMutationToolIndex(toolsUsed: ToolInvocation[]): number | null {
-  for (const [index, tool] of toolsUsed.entries()) {
-    if (tool.tool === "Write" || tool.tool === "Edit") {
-      return index;
-    }
-
-    if (tool.tool === "Bash" && extractBashCommands(tool.input).some(isLikelyMutatingBashCommand)) {
-      return index;
-    }
-  }
-
-  return null;
-}
-
-function isLikelyMutatingBashCommand(command: string): boolean {
-  return (
-    /\b(?:mkdir|touch|rm|mv|cp|install|tee)\b/.test(command) ||
-    /\b(?:cat|echo|printf)\b.*(?:>|>>|\|\s*tee\b)/.test(command) ||
-    /\bsed\s+-i\b/.test(command) ||
-    /\bperl\s+-pi\b/.test(command) ||
-    /\bwmill\b/.test(command)
-  );
-}
-
-function pushUnique(values: string[], value: string): void {
-  if (!values.includes(value)) {
-    values.push(value);
-  }
-}

ai_evals/adapters/frontend/backendPreview.test.ts

@@ -1,244 +0,0 @@
-import { afterEach, describe, expect, it } from 'bun:test'
-import type { BackendValidationSettings } from '../../core/backendValidation'
-import { BackendPreviewClient } from './backendPreview'
-
-const ORIGINAL_FETCH = globalThis.fetch
-
-afterEach(() => {
-	globalThis.fetch = ORIGINAL_FETCH
-})
-
-describe('BackendPreviewClient', () => {
-	it('updates an existing seeded script on path conflict and waits for deployment', async () => {
-		const requests: Array<{ url: string; init?: RequestInit }> = []
-		globalThis.fetch = mockFetch(
-			requests,
-			textResponse(200, 'token'),
-			textResponse(200, ''),
-			textResponse(400, 'Path conflict for f/evals/add_two_numbers with non-archived hash 123'),
-			jsonResponse(200, { hash: '123' }),
-			textResponse(200, '456'),
-			jsonResponse(200, { lock: 'script.lock', lock_error_logs: null })
-		)
-
-		const client = new BackendPreviewClient(
-			buildSettings({ baseUrl: 'http://backend.test/script-upsert' })
-		)
-
-		await client.createScript({
-			workspaceId: 'test',
-			path: 'f/evals/add_two_numbers',
-			summary: 'Add two numbers',
-			content: 'export async function main(a: number, b: number) { return a + b }',
-			language: 'bun'
-		})
-
-		expect(requests.map((entry) => entry.url)).toEqual([
-			'http://backend.test/script-upsert/api/auth/login',
-			'http://backend.test/script-upsert/api/w/test/folders/create',
-			'http://backend.test/script-upsert/api/w/test/scripts/create',
-			'http://backend.test/script-upsert/api/w/test/scripts/get/p/f/evals/add_two_numbers',
-			'http://backend.test/script-upsert/api/w/test/scripts/create',
-			'http://backend.test/script-upsert/api/w/test/scripts/deployment_status/h/456'
-		])
-
-		const updateRequest = requests[4]
-		expect(updateRequest.init?.method).toBe('POST')
-		expect(JSON.parse(String(updateRequest.init?.body))).toMatchObject({
-			path: 'f/evals/add_two_numbers',
-			parent_hash: '123',
-			language: 'bun'
-		})
-	})
-
-	it('updates an existing seeded flow on create conflict', async () => {
-		const requests: Array<{ url: string; init?: RequestInit }> = []
-		globalThis.fetch = mockFetch(
-			requests,
-			textResponse(200, 'token'),
-			textResponse(200, ''),
-			textResponse(400, 'Flow f/evals/add_numbers_flow already exists'),
-			textResponse(200, '')
-		)
-
-		const client = new BackendPreviewClient(
-			buildSettings({ baseUrl: 'http://backend.test/flow-upsert' })
-		)
-
-		await client.createFlow({
-			workspaceId: 'test',
-			path: 'f/evals/add_numbers_flow',
-			summary: 'Add numbers',
-			value: { modules: [] }
-		})
-
-		expect(requests.map((entry) => entry.url)).toEqual([
-			'http://backend.test/flow-upsert/api/auth/login',
-			'http://backend.test/flow-upsert/api/w/test/folders/create',
-			'http://backend.test/flow-upsert/api/w/test/flows/create',
-			'http://backend.test/flow-upsert/api/w/test/flows/update/f/evals/add_numbers_flow'
-		])
-
-		const updateRequest = requests[3]
-		expect(updateRequest.init?.method).toBe('POST')
-		expect(JSON.parse(String(updateRequest.init?.body))).toMatchObject({
-			path: 'f/evals/add_numbers_flow',
-			value: { modules: [] }
-		})
-	})
-
-	it('serializes shared-workspace validations inside the overridden workspace', async () => {
-		globalThis.fetch = async (input) => {
-			const url = String(input)
-			if (url.endsWith('/api/auth/login')) {
-				return textResponse(200, 'token')
-			}
-			if (url.endsWith('/api/workspaces/exists')) {
-				return textResponse(200, 'true')
-			}
-			if (url.endsWith('/api/w/shared-preview/flows/list_paths')) {
-				return jsonResponse(200, [])
-			}
-			if (url.endsWith('/api/w/shared-preview/scripts/list_paths')) {
-				return jsonResponse(200, [])
-			}
-			throw new Error(`Unexpected fetch: ${url}`)
-		}
-
-		const client = new BackendPreviewClient(
-			buildSettings({
-				baseUrl: 'http://backend.test/shared-lock',
-				workspaceOverride: 'shared-preview'
-			})
-		)
-
-		const order: string[] = []
-		let releaseFirst: (() => void) | undefined
-		let notifyFirstStart: (() => void) | undefined
-		const firstStarted = new Promise<void>((resolve) => {
-			notifyFirstStart = resolve
-		})
-
-		const first = client.withWorkspace('flow-test1', 1, async () => {
-			order.push('first:start')
-			notifyFirstStart?.()
-			await new Promise<void>((resolve) => {
-				releaseFirst = resolve
-			})
-			order.push('first:end')
-		})
-
-		const second = client.withWorkspace('flow-test2', 1, async () => {
-			order.push('second:start')
-			order.push('second:end')
-		})
-
-		await firstStarted
-		expect(order).toEqual(['first:start'])
-
-		releaseFirst?.()
-		await Promise.all([first, second])
-
-		expect(order).toEqual(['first:start', 'first:end', 'second:start', 'second:end'])
-	})
-
-	it('clears managed shared-workspace assets before preview runs', async () => {
-		const requests: Array<{ url: string; init?: RequestInit }> = []
-		globalThis.fetch = mockFetch(
-			requests,
-			textResponse(200, 'token'),
-			textResponse(200, 'true'),
-			jsonResponse(200, ['f/evals/old_subflow', 'u/admin/keep_flow']),
-			textResponse(200, ''),
-			jsonResponse(200, ['f/evals/old_script', 'f/shared/keep_script']),
-			textResponse(200, '')
-		)
-
-		const client = new BackendPreviewClient(
-			buildSettings({
-				baseUrl: 'http://backend.test/shared-cleanup',
-				workspaceOverride: 'shared-preview'
-			})
-		)
-
-		await client.withWorkspace('flow-test1', 1, async () => undefined)
-
-		expect(requests.map((entry) => entry.url)).toEqual([
-			'http://backend.test/shared-cleanup/api/auth/login',
-			'http://backend.test/shared-cleanup/api/workspaces/exists',
-			'http://backend.test/shared-cleanup/api/w/shared-preview/flows/list_paths',
-			'http://backend.test/shared-cleanup/api/w/shared-preview/flows/delete/f/evals/old_subflow',
-			'http://backend.test/shared-cleanup/api/w/shared-preview/scripts/list_paths',
-			'http://backend.test/shared-cleanup/api/w/shared-preview/scripts/delete/p/f/evals/old_script'
-		])
-	})
-
-	it('retries login after a cached login failure', async () => {
-		const requests: Array<{ url: string; init?: RequestInit }> = []
-		globalThis.fetch = mockFetch(
-			requests,
-			textResponse(503, 'backend starting'),
-			textResponse(200, 'token'),
-			textResponse(200, 'true'),
-			jsonResponse(200, []),
-			jsonResponse(200, [])
-		)
-
-		const client = new BackendPreviewClient(
-			buildSettings({
-				baseUrl: 'http://backend.test/login-retry',
-				workspaceOverride: 'shared-preview'
-			})
-		)
-
-		await expect(client.withWorkspace('flow-test1', 1, async () => undefined)).rejects.toThrow(
-			'login for backend validation failed'
-		)
-		await expect(client.withWorkspace('flow-test1', 1, async () => 'ok')).resolves.toBe('ok')
-
-		expect(
-			requests.filter((entry) => entry.url === 'http://backend.test/login-retry/api/auth/login')
-		).toHaveLength(2)
-	})
-})
-
-function buildSettings(
-	overrides: Partial<BackendValidationSettings> = {}
-): BackendValidationSettings {
-	return {
-		mode: 'preview',
-		baseUrl: 'http://backend.test/default',
-		email: 'admin@windmill.dev',
-		password: 'changeme',
-		pollIntervalMs: 1,
-		maxWaitMs: 50,
-		...overrides
-	}
-}
-
-function mockFetch(
-	requests: Array<{ url: string; init?: RequestInit }>,
-	...responses: Response[]
-): typeof fetch {
-	const queue = [...responses]
-	return async (input, init) => {
-		const url = String(input)
-		requests.push({ url, init })
-		const next = queue.shift()
-		if (!next) {
-			throw new Error(`Unexpected fetch: ${url}`)
-		}
-		return next
-	}
-}
-
-function jsonResponse(status: number, body: unknown): Response {
-	return new Response(JSON.stringify(body), {
-		status,
-		headers: { 'Content-Type': 'application/json' }
-	})
-}
-
-function textResponse(status: number, body: string): Response {
-	return new Response(body, { status })
-}

ai_evals/adapters/frontend/backendPreview.ts

@@ -1,503 +0,0 @@
-import { randomUUID } from 'node:crypto'
-import type { BackendValidationSettings } from '../../core/backendValidation'
-
-interface CompletedJobResultMaybe {
-	completed: boolean
-	result: unknown
-	success?: boolean
-	started?: boolean
-}
-
-interface ScriptDeploymentStatus {
-	lock?: unknown
-	lock_error_logs?: string | null
-}
-
-export interface CompletedPreviewJob {
-	id: string
-	success: boolean
-	result: unknown
-	logs?: string | null
-	raw: Record<string, unknown>
-}
-
-const tokenCache = new Map<string, Promise<string>>()
-const sharedWorkspaceQueue = new Map<string, Promise<void>>()
-const managedSharedWorkspacePrefixes = ['f/evals/']
-const DEFAULT_WORKSPACE_PREFIX = 'ai-evals'
-
-export class BackendPreviewClient {
-	constructor(private readonly settings: BackendValidationSettings) {}
-
-	async withWorkspace<T>(
-		caseId: string,
-		attempt: number,
-		body: (workspaceId: string) => Promise<T>
-	): Promise<T> {
-		const workspaceId =
-			this.settings.workspaceOverride ??
-			buildWorkspaceId(caseId, attempt)
-
-		const run = async () => {
-			await this.ensureWorkspace(workspaceId)
-			if (this.settings.workspaceOverride) {
-				await this.clearManagedSharedWorkspaceAssets(workspaceId)
-			}
-
-			try {
-				return await body(workspaceId)
-			} finally {
-				if (!this.settings.workspaceOverride) {
-					await this.deleteWorkspace(workspaceId).catch(() => undefined)
-				}
-			}
-		}
-
-		if (this.settings.workspaceOverride) {
-			return await withSharedWorkspaceLock(workspaceId, run)
-		}
-
-		return await run()
-	}
-
-	async createScript(input: {
-		workspaceId: string
-		path: string
-		summary: string
-		description?: string
-		schema?: Record<string, unknown>
-		content: string
-		language: string
-	}): Promise<void> {
-		await this.ensureFolderForPath(input.workspaceId, input.path)
-
-		const payload = {
-			path: input.path,
-			summary: input.summary,
-			description: input.description ?? '',
-			content: input.content,
-			schema: input.schema ?? { type: 'object', properties: {}, required: [] },
-			is_template: false,
-			language: input.language,
-			kind: 'script'
-		}
-
-		const response = await this.request(`/w/${encodeURIComponent(input.workspaceId)}/scripts/create`, {
-			method: 'POST',
-			headers: { 'Content-Type': 'application/json' },
-			body: JSON.stringify(payload)
-		})
-
-		if (response.ok) {
-			await this.waitForScriptDeployment(input.workspaceId, input.path, (await response.text()).trim())
-			return
-		}
-
-		const message = await response.text()
-		if (!isConflictMessage(message)) {
-			throw new Error(`create script ${input.path} failed: ${response.status} ${response.statusText} - ${message}`)
-		}
-
-		const currentScript = await this.getScriptByPath(input.workspaceId, input.path)
-		const currentHash = readStringField(currentScript, 'hash', `script ${input.path}`)
-		const updateResponse = await this.request(
-			`/w/${encodeURIComponent(input.workspaceId)}/scripts/create`,
-			{
-				method: 'POST',
-				headers: { 'Content-Type': 'application/json' },
-				body: JSON.stringify({
-					...payload,
-					parent_hash: currentHash
-				})
-			}
-		)
-		await expectOk(updateResponse, `update script ${input.path}`)
-		await this.waitForScriptDeployment(input.workspaceId, input.path, (await updateResponse.text()).trim())
-	}
-
-	async createFlow(input: {
-		workspaceId: string
-		path: string
-		summary: string
-		description?: string
-		schema?: Record<string, unknown>
-		value: Record<string, unknown>
-	}): Promise<void> {
-		await this.ensureFolderForPath(input.workspaceId, input.path)
-
-		const payload = {
-			path: input.path,
-			summary: input.summary,
-			description: input.description ?? '',
-			schema: input.schema ?? { type: 'object', properties: {}, required: [] },
-			value: input.value
-		}
-
-		const response = await this.request(`/w/${encodeURIComponent(input.workspaceId)}/flows/create`, {
-			method: 'POST',
-			headers: { 'Content-Type': 'application/json' },
-			body: JSON.stringify(payload)
-		})
-
-		if (response.ok) {
-			return
-		}
-
-		const message = await response.text()
-		if (!isConflictMessage(message)) {
-			throw new Error(`create flow ${input.path} failed: ${response.status} ${response.statusText} - ${message}`)
-		}
-
-		const updateResponse = await this.request(
-			`/w/${encodeURIComponent(input.workspaceId)}/flows/update/${input.path}`,
-			{
-				method: 'POST',
-				headers: { 'Content-Type': 'application/json' },
-				body: JSON.stringify(payload)
-			}
-		)
-		await expectOk(updateResponse, `update flow ${input.path}`)
-	}
-
-	async runScriptPreview(input: {
-		workspaceId: string
-		content: string
-		args: Record<string, unknown>
-		language: string
-		path?: string
-		timeoutSeconds?: number
-	}): Promise<CompletedPreviewJob> {
-		const response = await this.request(
-			withQuery(`/w/${encodeURIComponent(input.workspaceId)}/jobs/run/preview`, {
-				timeout: input.timeoutSeconds
-			}),
-			{
-				method: 'POST',
-				headers: { 'Content-Type': 'application/json' },
-				body: JSON.stringify({
-					content: input.content,
-					args: input.args,
-					language: input.language,
-					path: input.path
-				})
-			}
-		)
-
-		await expectOk(response, 'start script preview')
-		const jobId = (await response.text()).trim()
-		return await this.waitForCompletedJob(input.workspaceId, jobId)
-	}
-
-	async runFlowPreview(input: {
-		workspaceId: string
-		value: Record<string, unknown>
-		args: Record<string, unknown>
-		timeoutSeconds?: number
-		path?: string
-	}): Promise<CompletedPreviewJob> {
-		const response = await this.request(
-			withQuery(`/w/${encodeURIComponent(input.workspaceId)}/jobs/run/preview_flow`, {
-				timeout: input.timeoutSeconds
-			}),
-			{
-				method: 'POST',
-				headers: { 'Content-Type': 'application/json' },
-				body: JSON.stringify({
-					value: input.value,
-					args: input.args,
-					path: input.path
-				})
-			}
-		)
-
-		await expectOk(response, 'start flow preview')
-		const jobId = (await response.text()).trim()
-		return await this.waitForCompletedJob(input.workspaceId, jobId)
-	}
-
-	private async ensureWorkspace(workspaceId: string): Promise<void> {
-		const existsResponse = await this.request('/workspaces/exists', {
-			method: 'POST',
-			headers: { 'Content-Type': 'application/json' },
-			body: JSON.stringify({ id: workspaceId })
-		})
-		await expectOk(existsResponse, `check workspace ${workspaceId}`)
-
-		if ((await existsResponse.text()).trim() === 'true') {
-			return
-		}
-
-		const createResponse = await this.request('/workspaces/create', {
-			method: 'POST',
-			headers: { 'Content-Type': 'application/json' },
-			body: JSON.stringify({ id: workspaceId, name: workspaceId })
-		})
-		try {
-			await expectOk(createResponse, `create workspace ${workspaceId}`)
-		} catch (error) {
-			const message = error instanceof Error ? error.message : String(error)
-			if (message.includes('maximum number of workspaces')) {
-				throw new Error(
-					`${message}. Reuse an existing workspace with WMILL_AI_EVAL_BACKEND_WORKSPACE=<workspace-id>.`
-				)
-			}
-			throw error
-		}
-	}
-
-	private async deleteWorkspace(workspaceId: string): Promise<void> {
-		const response = await this.request(`/workspaces/delete/${encodeURIComponent(workspaceId)}`, {
-			method: 'DELETE'
-		})
-		await expectOk(response, `delete workspace ${workspaceId}`)
-	}
-
-	private async ensureFolderForPath(workspaceId: string, path: string): Promise<void> {
-		const folderName = extractFolderName(path)
-		if (!folderName) {
-			return
-		}
-
-		const response = await this.request(`/w/${encodeURIComponent(workspaceId)}/folders/create`, {
-			method: 'POST',
-			headers: { 'Content-Type': 'application/json' },
-			body: JSON.stringify({ name: folderName })
-		})
-
-		if (response.ok) {
-			return
-		}
-
-		const message = await response.text()
-		if (!message.toLowerCase().includes('already exists')) {
-			throw new Error(`Failed to create folder ${folderName}: ${message}`)
-		}
-	}
-
-	private async waitForCompletedJob(
-		workspaceId: string,
-		jobId: string
-	): Promise<CompletedPreviewJob> {
-		const deadline = Date.now() + this.settings.maxWaitMs
-
-		while (Date.now() < deadline) {
-			const maybeResponse = await this.request(
-				`/w/${encodeURIComponent(workspaceId)}/jobs_u/completed/get_result_maybe/${encodeURIComponent(jobId)}?get_started=false`
-			)
-			await expectOk(maybeResponse, `poll job ${jobId}`)
-			const maybeResult = (await maybeResponse.json()) as CompletedJobResultMaybe
-
-			if (maybeResult.completed) {
-				const completedResponse = await this.request(
-					`/w/${encodeURIComponent(workspaceId)}/jobs_u/completed/get/${encodeURIComponent(jobId)}`
-				)
-				await expectOk(completedResponse, `get completed job ${jobId}`)
-				const completedJob = (await completedResponse.json()) as Record<string, unknown>
-				return {
-					id: jobId,
-					success: Boolean(maybeResult.success),
-					result: maybeResult.result,
-					logs:
-						typeof completedJob.logs === 'string' || completedJob.logs === null
-							? (completedJob.logs as string | null)
-							: null,
-					raw: completedJob
-				}
-			}
-
-			await new Promise((resolve) => setTimeout(resolve, this.settings.pollIntervalMs))
-		}
-
-		throw new Error(`Timed out waiting for preview job ${jobId} to complete`)
-	}
-
-	private async getScriptByPath(workspaceId: string, path: string): Promise<Record<string, unknown>> {
-		const response = await this.request(`/w/${encodeURIComponent(workspaceId)}/scripts/get/p/${path}`)
-		await expectOk(response, `get script ${path}`)
-		return (await response.json()) as Record<string, unknown>
-	}
-
-	private async clearManagedSharedWorkspaceAssets(workspaceId: string): Promise<void> {
-		const flowPaths = await this.listFlowPaths(workspaceId)
-		for (const path of flowPaths.filter(isManagedSharedWorkspacePath)) {
-			await this.deleteFlowByPath(workspaceId, path)
-		}
-
-		const scriptPaths = await this.listScriptPaths(workspaceId)
-		for (const path of scriptPaths.filter(isManagedSharedWorkspacePath)) {
-			await this.deleteScriptByPath(workspaceId, path)
-		}
-	}
-
-	private async listFlowPaths(workspaceId: string): Promise<string[]> {
-		const response = await this.request(`/w/${encodeURIComponent(workspaceId)}/flows/list_paths`)
-		await expectOk(response, `list flows in workspace ${workspaceId}`)
-		return await response.json()
-	}
-
-	private async listScriptPaths(workspaceId: string): Promise<string[]> {
-		const response = await this.request(`/w/${encodeURIComponent(workspaceId)}/scripts/list_paths`)
-		await expectOk(response, `list scripts in workspace ${workspaceId}`)
-		return await response.json()
-	}
-
-	private async deleteFlowByPath(workspaceId: string, path: string): Promise<void> {
-		const response = await this.request(`/w/${encodeURIComponent(workspaceId)}/flows/delete/${path}`, {
-			method: 'DELETE'
-		})
-		await expectOk(response, `delete flow ${path}`)
-	}
-
-	private async deleteScriptByPath(workspaceId: string, path: string): Promise<void> {
-		const response = await this.request(`/w/${encodeURIComponent(workspaceId)}/scripts/delete/p/${path}`, {
-			method: 'POST'
-		})
-		await expectOk(response, `delete script ${path}`)
-	}
-
-	private async waitForScriptDeployment(
-		workspaceId: string,
-		path: string,
-		hash: string
-	): Promise<void> {
-		const deadline = Date.now() + this.settings.maxWaitMs
-
-		while (Date.now() < deadline) {
-			const response = await this.request(
-				`/w/${encodeURIComponent(workspaceId)}/scripts/deployment_status/h/${encodeURIComponent(hash)}`
-			)
-			await expectOk(response, `check deployment status for script ${path}`)
-			const deployment = (await response.json()) as ScriptDeploymentStatus
-			if (deployment.lock != null) {
-				return
-			}
-			if (deployment.lock_error_logs) {
-				throw new Error(`Script deployment failed for ${path}: ${deployment.lock_error_logs}`)
-			}
-			await new Promise((resolve) => setTimeout(resolve, this.settings.pollIntervalMs))
-		}
-
-		throw new Error(`Timed out waiting for script ${path} (${hash}) to deploy`)
-	}
-
-	private async request(path: string, init?: RequestInit): Promise<Response> {
-		const token = await this.getToken()
-		return await fetch(`${this.settings.baseUrl}/api${path}`, {
-			...init,
-			headers: {
-				Authorization: `Bearer ${token}`,
-				...(init?.headers ?? {})
-			}
-		})
-	}
-
-	private async getToken(): Promise<string> {
-		const cacheKey = `${this.settings.baseUrl}|${this.settings.email}`
-		let tokenPromise = tokenCache.get(cacheKey)
-		if (!tokenPromise) {
-			tokenPromise = this.login().catch((error) => {
-				if (tokenCache.get(cacheKey) === tokenPromise) {
-					tokenCache.delete(cacheKey)
-				}
-				throw error
-			})
-			tokenCache.set(cacheKey, tokenPromise)
-		}
-		return await tokenPromise
-	}
-
-	private async login(): Promise<string> {
-		const response = await fetch(`${this.settings.baseUrl}/api/auth/login`, {
-			method: 'POST',
-			headers: { 'Content-Type': 'application/json' },
-			body: JSON.stringify({
-				email: this.settings.email,
-				password: this.settings.password
-			})
-		})
-		await expectOk(response, 'login for backend validation')
-		return (await response.text()).trim()
-	}
-}
-
-async function withSharedWorkspaceLock<T>(workspaceId: string, body: () => Promise<T>): Promise<T> {
-	const previous = sharedWorkspaceQueue.get(workspaceId) ?? Promise.resolve()
-	let releaseCurrent: (() => void) | undefined
-	const current = new Promise<void>((resolve) => {
-		releaseCurrent = resolve
-	})
-	const tail = previous.catch(() => undefined).then(() => current)
-	sharedWorkspaceQueue.set(workspaceId, tail)
-
-	await previous.catch(() => undefined)
-
-	try {
-		return await body()
-	} finally {
-		releaseCurrent?.()
-		if (sharedWorkspaceQueue.get(workspaceId) === tail) {
-			sharedWorkspaceQueue.delete(workspaceId)
-		}
-	}
-}
-
-function buildWorkspaceId(caseId: string, attempt: number): string {
-	const caseSlug = caseId
-		.toLowerCase()
-		.replace(/[^a-z0-9-]+/g, '-')
-		.replace(/^-+|-+$/g, '')
-		.slice(0, 30)
-	const suffix = randomUUID().slice(0, 8)
-	return `${DEFAULT_WORKSPACE_PREFIX}-${caseSlug || 'case'}-a${attempt}-${suffix}`
-}
-
-function extractFolderName(path: string): string | null {
-	if (!path.startsWith('f/')) {
-		return null
-	}
-	const segments = path.split('/').slice(1, -1)
-	return segments.length > 0 ? segments.join('/') : null
-}
-
-function withQuery(
-	path: string,
-	params: Record<string, string | number | undefined>
-): string {
-	const query = new URLSearchParams()
-	for (const [key, value] of Object.entries(params)) {
-		if (value === undefined) {
-			continue
-		}
-		query.set(key, String(value))
-	}
-	const suffix = query.toString()
-	return suffix ? `${path}?${suffix}` : path
-}
-
-async function expectOk(response: Response, context: string): Promise<void> {
-	if (response.ok) {
-		return
-	}
-	throw new Error(`${context} failed: ${response.status} ${response.statusText} - ${await response.text()}`)
-}
-
-function readStringField(
-	value: Record<string, unknown>,
-	field: string,
-	context: string
-): string {
-	const candidate = value[field]
-	if (typeof candidate === 'string' && candidate.length > 0) {
-		return candidate
-	}
-	throw new Error(`${context} is missing string field ${field}`)
-}
-
-function isConflictMessage(message: string): boolean {
-	const normalized = message.toLowerCase()
-	return normalized.includes('already exists') || normalized.includes('path conflict')
-}
-
-function isManagedSharedWorkspacePath(path: string): boolean {
-	return managedSharedWorkspacePrefixes.some((prefix) => path.startsWith(prefix))
-}