mirror of
https://github.com/stablyai/orca.git
synced 2026-11-13 09:27:29 +00:00
hide-codex-error-msg
15
Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
  
|
f13a67a7c4 |
fix(skills): stop the orchestration stub repeating its description (#37850)
The orchestration stub body opened with a near-verbatim copy of its front matter description plus a paragraph on why the guide lives in the binary. Give it the one-line opening every other stub uses and keep its only unique rule: never substitute a non-Orca subagent tool for Orca orchestration. The description is unchanged since it drives skill routing. Regenerated skills/orchestration/SKILL.md, the bundle manifest, bundled sources, and the released-digest ledger (adds the new stub digest only). Fixes #37849 Co-authored-by: Michael <mhoeykens@wattvibe.eu> Co-authored-by: Jinwoo-H <jinwoo0825@gmail.com> |
||
 Kelvin Amoaba
|
7f885c19a6 |
fix(skills): rename computer use to orca-computer-use (#37659)
* fix(skills): rename computer use to orca-computer-use * test(skills): resolve rename fixture after main cleanup |
||
 Neil
|
b17d408416 |
Plugin dev loop: TS workers, hot reload with state, orca plugin dev watch, authoring skill (#36982)
* feat(plugins): build-less TypeScript workers A worker's main may be .ts/.mts. The plugin host installs module.registerHooks scoped to the plugin folder: .ts files are transpiled with the frames' sucrase options (lines kept, inline source map, source maps on), relative imports fall back to the frames' candidate files, and a value import of the type-only 'orca/worker' fails with a clear error. Needs engines.orca >=1.4.246. * feat(plugins): components panels hot reload with their state - In dev folders a save swaps only the hidden guest: keyed memory atoms are handed to the new guest (ui.retain → ui.hydrate memory), dropped when an atom's version changes. - A save that fails to load keeps the last good render and shows an inline file:line:col error banner; the next good save clears it. A subtle Reloaded mark follows a reload. - Transpile and import errors carry line and column; dev worker restarts keep storage. * feat(plugins): orca plugin dev watch One command for the inner loop: refreshes .orca/types, dev-loads the folder, follows its frame and worker logs (plugins.getLogs gains an optional cursor), prints reloads (optional devChangeCount on list entries) and status changes, and unloads on Ctrl-C. --json streams NDJSON events for agents. * feat(skills): orca-plugin-authoring bundled skill guide A guide for agents to build Orca plugins end to end (decision table, build-less rules, the dev watch loop, test/check, validate, release-entry, approvals, gotchas) with references generated from the component catalog, host API table, event schemas, manifest schema and CLI specs, checked for drift. Fixes the components template's Reset to set(count, 0). * feat(plugins): orca plugin dev watch --check; test kit details in the authoring skill - --check runs orca plugin check after the load and after each reload (one at a time; saves made meanwhile fold into one rerun) and prints each panel's render or errors inline, as a `check` event with --json. Local folders only, since it runs the plugin's code here. - The orca-plugin-authoring guide's Test and check section now documents orca plugin test and check from #36969 exactly (JSON results, exit codes, test placement, the orca/testing kit and its common mistakes), and the loop uses dev watch --check. * refactor(plugins): move the components renderer's timing and helper constants to their own module |
||
 Jinwoo Hong
|
8138232c4f |
Print concise orchestration CLI output for agents, with --full for the full record (#36498)
Orchestration commands print small flat fields in text and --json by default; --full prints the previous output. Failures exit non-zero with error: <code>: <message> on stderr. |
||
|
Jinwoo Hong
|
a1f7f97671 |
Orchestration v3: coordinators end their turn and Orca wakes them (#36137)
* fix(orchestration): keep a terminal's mail from being consumed by another terminal, and let settled structured workers take further tasks A plain orchestration check consumed the named terminal's mail whatever terminal ran it, so a coordinator checking a worker's handle took the worker's message. A consuming check from a terminal whose own live pane is a different terminal's is now refused; --peek and --all still read it without consuming. worker-start --terminal and dispatch --inject routed a structured worker (structworker_ handle) as a terminal and failed with terminal_unsupported_for_agent_session. Any agent-session party now takes its task as a send into its session, while a structured worker keeps the pane and process authority its first Dispatch had. * fix(orchestration): fence check by live terminal incl. acks; defer injecting structured workers Review fixes for the mail fence and structured worker reuse: - The check fence compares the caller's and the mailbox's live terminal process instead of recorded panes, so exited terminals keep the old behaviour and two processes sharing a recorded pane are told apart. It also covers --ack in every read mode. The caller lookup is shared with the lifecycle caller fence. - dispatch --inject to a structured worker stays refused as before: an injected Dispatch takes no custody, so releasing the first Dispatch could close a session working on the second. worker-start --terminal reuse transfers custody and stays supported. - worker-start decides pane-less chat authority from the resolved party and passes the session down, so delivery and next commands no longer re-derive it from the address spelling. - Receipts name an agent session, not a chat. * fix(orchestration): an exited caller's env pane names no caller The caller lookup fell back to the env pane key whenever the env handle was not live, so an exited terminal's pane resolved to whatever terminal now occupies it, and a tmux worker's own check was refused. The pane fallback now applies only when the env handle is not proven exited. * fix(terminal): keep surviving terminals streaming to main after a restart Main now attaches every live local daemon session it adopts at startup, not only ones a window pane or remote view is showing, and re-attaches after a daemon endpoint drop. The attach snapshot seeds main's screen, tail and title, and an attached-but-silent agent gets an output clock so readiness can see it has gone quiet. Waits on a survivor main could not attach fail at once instead of timing out. * fix(terminal): restore only pane-less streams after endpoint loss; one attach state - Endpoint loss re-attaches in main only sessions no renderer view reports; a pane restores its own. - The attach hide skips views mid-attach and leaves a pane's own hidden claim alone on failure. - One per-session stream state (attaching, attached, lost) replaces three sets and is forgotten on exit. - Waits retry the attach before failing, including after a failed re-attach. - The output clock starts wherever a reattach result is seeded, so pane reattaches get it too. * refactor(terminal): attach survivors at inventory adoption only; waits on unattached survivors fail at once; one ptyId output clock * test(e2e): scope the restart attach spec's terminal listing to the unopened workspace * fix(terminal): narrow survivor attach: waits behave as on main, one quiet clock per generation, keep main's hidden mark on failed attach * Teach the orchestration guide when to hand off and to end the turn after dispatch * Rewrap a messaging reference line * Keep the original placement default and skill trigger; keep the accuracy fixes * Point DAG retries at --retry-of and drop the last blocking-wait instruction * Check the served orchestration guide by its new worker heading * Apply council review: best-effort wake, full completion rule, restored safety rules, fixed scope and settled wording * Say when the orchestration skill applies, and prefer own subagents otherwise * Name the polling, steering, checkout, and worker fan-out habits to avoid * fix(orchestration): wake a coordinator whose turn ended while its program still reports working Claude Code keeps its OSC 7501 report on working while an agent-teams teammate is alive, even after the lead's own turn ends. The mailbox gate, idle edges and tui-idle all read that report, so worker mail was stored and never typed. A hook turn end newer than the program's working report now counts as rest; hook rows re-classify the PTY so the existing idle edge delivers. Agent Teams panes now join their claude hook rows. * fix(orchestration): compare the lead's own turn end, not the row's receipt A child's hook restamps the row's receivedAt while the lead's turn is still the old one, which could mark rest under a newer program working report. Use the lead's turn-end stamp on this runtime's clock, after the latest input. * feat(orchestration): a consuming check marks the batch it returns as handled * Say why watch scripts and open waits block the wake * feat(orchestration): release settled workers automatically after a grace period * fix(orchestration): use checked row reads for delivery cursors * fix(orchestration): trust a live own-process turn end instead of comparing clocks Drop programActivitySince and the turn-end clock mapping: replays and SSH host clocks made the comparison unreliable. A Claude lead counts as at rest under a program working report when its hook turn end is a live transition from this pane's host and launch, past the existing input and respawn floors. * test(orchestration): drop cast in ack no-op check * test(orchestration): seed legacy outstanding batch for held chat nudges * test(orchestration): built CLI check hands its batch over * test(orchestration): annotate unchanged casts and stub the keep item in tab menu tests * fix(orchestration): let real input, explicit release and host releases converge with automatic release - Record worker takeover at the owning runtime when it accepts UI input, before the bytes reach the PTY. - Run remote attachment releases through the same per-resource single-flight as local ones. - Apply the confirmed-release home projection when host observation reports a host-side release, and carry the host deadline in fleet snapshots. - Prove restart recovery with a file-backed database reopened by a fresh runtime. * test(orchestration): classify the keep RPC's worker fields instead of counting methods * fix(orchestration): checked delivery reads and observer and recovery help * refactor(orchestration): end a check when it hands over a batch; wake rides the wait result * fix(orchestration): refuse an automatic close after real input, join remote releases, converge the home - An automatic close checks the terminal's last driving-input time right before closing and keeps a worker someone typed into after worker_done; typing itself does no orchestration database work. - Remote releases share the per-resource single-flight with local release, so a home worker-release joins the host timer's close. - Host fleet snapshots and federationShow report released and the pending deadline; the home applies the existing release projection inside its observation fence. - Real restart tests: a file-backed database reopened by a fresh runtime releases a missed deadline once and finishes interrupted requests. * fix(orchestration): receipt a shared wake batch on arrival and report the applied Run ack * fix(orchestration): deliver the wake to a working coordinator; agents queue mid-turn input The wake was typed only into a live idle pane, so a Claude lead whose program status stayed working (teammate alive) never got it. Agents queue or steer text typed mid-turn, so a pane takes mail when live idle or working; dialogs, permission prompts, exited agents, unsettled panes and cold-parked Enter still hold it. Drops the held-batch Enter deferral and the per-turn-end hook logic. * fix(orchestration): record each remote release request before joining, publish archives only through the conditional commit - Each remote release caller records its own request before joining a close under way, and the close decides on the current row, so an explicit release that overlaps the host timer clears the automatic deadline. - Remote capture publishes its archive through commitWorkerTerminalArchiveForRelease, so a retain during capture leaves no stale archive for worker-read. - Remote worker-show projects the fenced host deadline and its next action. - The input guard lives in its own module instead of the scheduler. * feat(orchestration): start a worker in one step; worker-start creates the coordinator's Run when none is bound worker-start --spec now creates and binds a Run for a root coordinator with no bound Run, through the same path run-create uses, and names the Run it created or joined in every receipt. Workers on a Dispatch keep starting helpers in their own Run and never get a new one. A new CLI against an older runtime runs run-create and retries; old CLIs need no change. * fix(orchestration): drop the older-runtime fallback; check the spec before a Run is created The CLI fallback matched a refusal text the client decorates, could not prove the caller was a root coordinator on an older runtime, and recovered an inner run-create under the worker-start command. Older runtimes keep their refusal and the guide says to run-create first there. The prompt budget check now runs before worker-start can create a Run. * Route report-back requests to coordinators, complete the steer command, and label the wake loop branches * Say workers are done only when worker_done confirms, wait for helpers to settle, and let bare-shell workers exit * Route report-back requests away from handoff and steer supervised workers through their mailbox * docs(orchestration): teach auto-handled check, retry-request and inbox recovery in the guides * Describe the mid-turn wake in the messaging reference * docs(orchestration): say terminal workers are released 2 minutes after worker_done - The guide's Completion section, the reuse recipe and the recovery reference now describe the automatic release and its autoReleaseAt deadline: reuse or worker-retain before it; explicit worker-release still works and is required for chat workers and workers without a deadline. - The worker preamble says the terminal stays available until Orca releases it, about 2 minutes after worker_done unless retained. - Regenerated the bundled guides. * feat(orchestration): receipts say when to end the turn * feat(orchestration): tell the coordinator when a worker stops without worker_done * test+docs: worker stop reports * fix(orchestration): judge the empty-check stop with execution-host verdicts * test: drop casts from worker stop tests * Say that report-back requests still mean end the turn * Say that report-back requests still mean end the turn * fix(orchestration): stop reports wait for children, keep their Dispatch, and relay intact * fix(orchestration): a stop belongs to the Dispatch active when it was seen; restored stamps announce nothing * fix(orchestration): never say to stop while a remote worker runs * Shrink the orchestration guide to the end-state workflow * fix(orchestration): stop reports quote the agent's last words, or the screen for errors and prompts * fix(orchestration): withhold the stop while chat failure mail is off * test: a clean finish quotes the agent's last words * fix(orchestration): stop asking workers for heartbeats Workers are no longer told to send periodic heartbeats. A heartbeat that still arrives from an older worker or peer only records last_heartbeat_at: it is stored already handled, so it never lands in a coordinator inbox and never wakes a coordinator or a check --wait. Removes the stale-heartbeat warning and the heartbeat interval plumbing that only fed the preamble. * fix(settings): remove the worker heartbeat controls The Worker heartbeats switch and Heartbeat interval field only changed the worker preamble text, which no longer mentions heartbeats. Saved values are dropped on load as retired settings. * fix(orchestration): wake a coordinator that reports program status but never sets a title Program status only re-classified a pane that had a title, so a titleless agent (agy) never got a live status and the wake gate refused it. * fix(orchestration): a new Run no longer cuts a coordinator off from its earlier Run * test(orchestration): drop heartbeat-batch cases the handled heartbeat row removed * fix(orchestration): redact capabilities from a stop report's quoted output * fix(orchestration): never wake a mailbox for a heartbeat One rule at the runtime notification entry replaces per-call-site read checks, so mutation replay and federated control-mail import stay silent for heartbeats too. * Shrink the orchestration references and worker preamble to the end-state workflow * Name the older-Orca bind refusal as worker-start prints it * Owner-fence test: a refused foreign ack changes nothing, whatever the worker's own check already handled * Integration: drop the duplicate consumingCheck after merging #34043 and #35293 * Integration: await the worker-start next receipt before naming its Run * Apply review round 1: drain to zero, release without auto-release, retry after stop, Run before a DAG * Apply review round 2: start ready DAG Tasks, block when no failure wake, replay sends by request ID * Apply review round 3: one receipt-based wait rule for starts and wakes * fix(orchestration): a worker stopped without reporting needs action, and stop keeps its output worker-list projected a stop-reported attempt as live, in_progress, no action. It now projects finished_unverified with a worker-read recovery until mail reaches the worker. worker-stop closed the terminal before anything archived it, so release reported the archive unavailable; stop now archives the output first. * Revert "Owner-fence test: a refused foreign ack changes nothing, whatever the worker's own check already handled" This reverts commit |
||
|
Neil
|
b35fb1ccfb |
fix(orchestration): route dispatched workers to the skill and stop heartbeat loops before worker_done (#34404)
The orchestration skill description only advertised coordinator work, so a dispatched worker never loaded it and built its heartbeat loop from the preamble alone. Neither the preamble nor the guide said to stop a background heartbeat loop before worker_done, so the loop outlived the task. Add worker-role triggers (heartbeats, ask, worker_done) to the description and stub, and tell workers in the preamble and guide to restart an expired heartbeat loop mid-task and stop it before worker_done. Fixes #26605 |
||
 
|
7ac2dcc7d4 |
feat(cli): let agents read and update Jira tickets safely (#28592)
Expose host-owned verified Jira reads and single-attempt writes through the CLI. Preserve site ambiguity checks, old-host capability refusal, atomic label changes, and uncertainty after failed writes. Refs #11986 and #26055. Co-authored-by: Dan Cieslak <dcieslak19973@users.noreply.github.com> Co-authored-by: kokbee <kok2209@gmail.com> |
||
 Jinwoo Hongandlifeodyssey
|
9af6a3d798 |
fix(cli): report a denied runtime connection instead of a dead Orca (#22341)
* fix(cli): report a denied runtime connection instead of a dead Orca Inside Codex's macOS Seatbelt sandbox, connect() on the runtime socket fails with EPERM. The CLI dropped the errno and reported "Could not connect ... Restart Orca", appended "Orca is not running. Run 'orca open' first.", and `orca status` answered ok:true with `starting` (its pid probe also gets EPERM). An agent following that advice restarts a healthy app, which cannot help. EPERM/EACCES on the metadata read or the socket/pipe connect now fails with a CLI-local `runtime_access_denied` error: ok:false, non-zero exit, operation/systemCode/processState:"unverifiable"/retryable:false and nextSteps that say to re-run with escalated permissions and not restart. CODEX_SANDBOX only picks the wording. `orca open` stops before launching. The status pid probe is unchanged: a refused or missing socket proves the caller reached the endpoint, so a later EPERM probe is another uid and keeps #20098's `starting`. Missing, refused, stale-pid and timeout paths are unchanged. Adapted from the diagnosis and tests in #20487 (and #19605, #13583). Co-authored-by: lifeodyssey <zhenjiazhou0127@outlook.com> * docs(skills): tell agents runtime_access_denied means escalate, not restart The shared CLI-resolution block told every bundled skill to run `orca open` when a command says Orca is not running. Add the counterpart for the new access-denied code so sandboxed agents re-run with escalated permissions instead of launching or restarting Orca. Regenerated stubs and manifest. * refactor(cli): classify only a denied runtime connect, with a leaner error A denied metadata read was never observed under a sandbox, and it turned an unreadable user-data path (the Linux launch contract's root-owned HOME) into runtime_access_denied instead of "Orca is not running". Keep metadata reads as on main and classify only the socket/pipe connect. One helper now maps a socket errno to the error or null; the error data keeps only systemCode and nextSteps. Tests drop cases already pinned by status.test.ts. * fix(cli): give not-running advice when a denied socket belongs to a dead Orca A crashed Orca leaves its metadata and socket file behind, and a sandbox denies the connect with EPERM before the CLI can see ECONNREFUSED. The sandbox still reports ESRCH for a gone pid, so a denied connect now probes the metadata pid and falls through to the ordinary unavailable path when the pid is proven gone. isProcessRunning moves to its own module so transport and status share it. * refactor(cli): inline the runtime_access_denied code like other CLI error codes --------- Co-authored-by: lifeodyssey <zhenjiazhou0127@outlook.com> |
||
|
Jinwoo Hong
|
fb322046e8 |
skills: rewrite and trim the seven non-orchestration guides (#19128)
* skills: rewrite the seven non-orchestration guides to one outcome-first standard
Every guide leads with Result / Done / Safe failure, states conditions instead of case lists, keeps one done bar and one autonomy envelope, and loads references at the point of use via `skills get <topic> --full`. orca-cli drops from 424 to 260 always-loaded lines with three references; orca-per-workspace-env from 794 to 397 with five.
Defects fixed in shipped guides: `emulator camera` (no such command), iOS `permissions` (backend refuses it), Android pane described as in development, `relayGracePeriodSeconds: 0` documented as immediate teardown (it is unbounded), doctor `ok: true` hiding `warn`, an SSH exemplar setting both `jumpHost` and `proxyCommand`, a provisioned-root fetch from `origin`, and the Linear unconfirmed-write rule keyed on four verbs when ten emit it.
The resolver ladder, placeholder rule, and older-binary fallback shared by every installable SKILL.md now come from one skill-stubs/_shared/cli-resolution.md fragment composed by the generator, which also bundles per-guide references into --full. New guards: every ORCA invocation and flag resolves against COMMAND_SPECS, descriptions carry no angle-bracket tokens, reference routing is checked both ways, and an always-loaded size ratchet (300 lines) that guides may leave but never join.
* skills: address review on the SSH recipe and the parity guard
- ssh-host create script: route the bootstrap ssh through the chosen jump host or proxy command, refuse both at once, use StrictHostKeyChecking=accept-new instead of a blind ssh-keyscan append, and pass gh_token/project_root/repo_url/repo_ref to the remote bash via printf %q so a quote in a value cannot break out of the command.
- per-workspace-env envelope: the step-10 workspace test the user asked for is no longer forbidden by the same paragraph.
- linear guides: name the full verb, ORCA linear list-issues.
- parity guard: a prefix reference such as ORCA linear --help or ORCA emulator --webcam now has its flags checked against every command under that prefix; only an exact path or an explicit ... was checked before.
* skills: tighten prose in the seven rewritten guides
Shorter outcome spines, one idea per sentence, no restated rationale after a rule. No rule, command, or pinned phrase changes; 47 net lines fewer across the guides and references.
* skills: route orca-cli and per-workspace-env gates through --reference
Both guides told agents to load --full at a gate because the per-reference
selector did not exist when they were written. Now that main serves
`skills get <topic> --reference references/<file>.md`, load only the
named file and keep --full as the fallback for an older CLI, matching the
orchestration kernel.
* skills: drop outcome-spine boilerplate from the CLI-wrapper guides
The Result/Done/Safe-failure preambles and Next Action closers restated
rules the body already carries. Agents stop fine without them, and for
a CLI wrapper the command surface is the guide. Keeps the one substantive
rule computer-use's Done block added (never report unverified as success)
inside Action Rules. orchestration and per-workspace-env keep theirs:
those are multi-step workflows where the done bar is load-bearing.
(cherry picked from commit
|
||
|
Jinwoo Hong
|
b8311d509a |
Revert "skills: rewrite the seven non-orchestration guides to one outcome-first standard (#18724)" (#19126)
This reverts commit
|
||
|
Jinwoo Hong
|
15d0f8aedf |
skills: rewrite the seven non-orchestration guides to one outcome-first standard (#18724)
<!-- orca-pr-loc -->
<!-- Programmatic LoC summary. Do not edit by hand; rewritten on every commit. -->
| | Files | Added | Deleted | Net |
| :--- | ---: | ---: | ---: | ---: |
| Test | 6 | $\color{#1a7f37}{\Huge{\mathbf{+}}}$544 | $\color{#cf222e}{\Huge{\mathbf{−}}}$49 | $\color{#1a7f37}{\Huge{\mathbf{+}}}$495 |
| Prod | 36 | $\color{#1a7f37}{\Huge{\mathbf{+}}}$1719 | $\color{#cf222e}{\Huge{\mathbf{−}}}$1703 | $\color{#1a7f37}{\Huge{\mathbf{+}}}$16 |
<!-- /orca-pr-loc -->
## ELI5
Orca ships eight skill guides that agents read before running the CLI. Seven of them (everything except `orchestration`, which #16904 rewrites) were command catalogs that had drifted from the binary. This PR rewrites them so an agent reads the outcome, the done bar, and the safe-failure rule first, loads reference material only at the step that needs it, and never sees a command or flag the installed CLI does not define.
## What changed
- **Seven guides rewritten** to one standard: outcome spine first (Result / Done / Safe failure), conditions instead of case lists, one done bar, one autonomy envelope, references loaded at the point of use via `skills get <topic> --full`, every runnable invocation spelled `ORCA`. `orca-cli` is 424→260 always-loaded lines with three references (browser, automations, publishing); `orca-per-workspace-env` is 794→397 with five (provider-vercel, ssh-host, docker-ssh, windows-scripts, failure-modes).
- **Defects fixed in shipped guides:** `emulator camera` (no such command), iOS `permissions` (backend refuses it), Android pane described as "in development" (shipped in June), `relayGracePeriodSeconds: 0` documented as immediate teardown (it is unbounded), doctor `ok: true` hiding `warn`, an SSH exemplar setting both `jumpHost` and `proxyCommand`, a provisioned-root fetch from `origin`, the Linear unconfirmed-write rule keyed on four verbs when ten emit it. Linear and emulator descriptions dropped embedded commands and angle-bracket placeholders (651→329, 732→404 chars).
- **Generator bundles references.** `skill-guides/<name>/references/*.md` is appended to `--full`; `skills get` help says compact by default, full with references.
- **Stubs single-authored.** The resolver ladder, placeholder rule, and older-binary fallback shared by all eight installable `SKILL.md` files come from one `skill-stubs/_shared/cli-resolution.md` fragment composed by the generator. Projections were byte-identical before the content fixes.
- **Guards:** every `ORCA <cmd>` and flag in every guide and reference resolves against `COMMAND_SPECS` (this found the camera defect); descriptions ≤1024 chars with no angle-bracket tokens; reference routing checked both directions; an always-loaded size ratchet (300 lines) that guides may leave but never join. `orchestration` (440 lines on main) is recorded as an exception until #16904 lands its kernel.
## Relationship to #16904
Split out of #16904 so that PR carries only the orchestration guide. On main, `terminal send` has no `--wait-submit` / `--retry-request` and the orchestration kernel still carries the resolver ladder and worktree-selector rule, so this branch pins `accepted: true` for handoff receipts and leaves the orchestration pins where main has them. The merge in either direction is mechanical: #16904 rebased on this becomes a one-file `orchestration.md` change plus dropping the two exceptions.
## Standard
Compound Engineering's portable skill-authoring guidance (outcome spine, conditions not cases, pinned fragile commands with an ordered hatch, references at point of use). NVIDIA SkillEvaluator Tier 1 (`schema,pii,license,quality,unicode,lint`) was run on every guide; its deterministic checks pass, its template nudges (Instructions/Examples sections, 50–150 char descriptions) do not apply to Orca's stub architecture and were not applied.
## Testing
- `pnpm typecheck:tsc:cli` clean; `check:code-quality:changed` and `check:react-doctor:changed` 0 findings
- `pnpm verify:bundled-skill-guides` and skill-bundle manifest verify clean
- vitest over `config/scripts`, `src/cli/skill-guide-cli-parity.test.ts`, `src/cli/skills.test.ts`, `src/cli/specs/skills.test.ts`, `src/cli/help.test.ts`, `src/main/skills`: 240 files / 2,019 pass
- Live smoke on the built CLI of every `skills get <topic>` and `--full`, every emulator, linear, and vm verb named in the guides, and every projection's resolver, GNOME warning, and bounded fallback (done on the #16904 branch before the split; the guide bodies are identical here except the send-receipt vocabulary noted above)
## Deferred product decisions
Merging `orca-emulator` and `orca-emulator-android` into one skill with a platform branch; collapsing `linear-tickets` to a guide alias; a `skills get --reference <name>` selector so a gate table can load one file; a fresh-agent routing eval before trimming the `orca-cli` (1,015 chars) and `orchestration` descriptions, whose quoted triggers each fixed a routing misroute.
|
||
|
Jinwoo Hong
|
06a607a1d7 |
feat(orchestration): make multi-agent workflows durable (#16904)
<!-- orca-pr-loc -->
<!-- Programmatic LoC summary. Do not edit by hand; rewritten on every commit. -->
| | Files | Added | Deleted | Net |
| :--- | ---: | ---: | ---: | ---: |
| Test | 225 | $\color{#1a7f37}{\Huge{\mathbf{+}}}$21666 | $\color{#cf222e}{\Huge{\mathbf{−}}}$2820 | $\color{#1a7f37}{\Huge{\mathbf{+}}}$18846 |
| Prod | 348 | $\color{#1a7f37}{\Huge{\mathbf{+}}}$17107 | $\color{#cf222e}{\Huge{\mathbf{−}}}$4706 | $\color{#1a7f37}{\Huge{\mathbf{+}}}$12401 |
<!-- /orca-pr-loc -->
## ELI5
Orca now treats orchestration like a durable control plane instead of inferring success from terminal keystrokes. Agents can tell whether a prompt was accepted or a turn started, replay an ambiguous request without sending twice, and recover coordinator mail after a crash. Completed workers can be inspected, released, or retained, and their panes no longer auto-resume as if the work were still running.
## What changed
- **Run receipts** from `run-create/use/current/show/list` are the row without routing plumbing (`home_database`, `coordinator_pane_key`) and without the duplicate `binding` object.
- **`terminal send` receipts are honest and idempotent.** `input_accepted` and `turn_started` are the only stages; `--wait-submit` observes without resending; `--retry-request <uuid>` replays the exact request against the same process incarnation. A transport timeout keeps the retry ID; only a different runtime answering strips it. Value-less or non-UUID `--retry-request` is rejected on the CLI and the SSH shim.
- **Mailbox delivery is committed before wakeup.** Pointer writes are staged in the DB before any PTY byte, replayed once after restart, and never emit a naked Enter. The watermark that parks concurrent deliveries is released with the DB reservation. Restart rescans pointer-pending and `dispatch:` mailboxes.
- **Lifecycle is a guarded transition graph** (`lifecycle-transition.ts`) with a table-driven test over every caller edge. Task reopen/overturn stays in the public contract. A PTY exit during `worker-stop` is the stop succeeding, not a failure.
- **Worker lifecycle CLI:** `worker-start` (`--spec` creates Task + attempt in one call), `worker-show`, `worker-read` (provider transcript first, bounded terminal fallback with a typed reason, local/WSL/SSH), `worker-stop`, `worker-abandon`, `worker-release`, `worker-retain`, `worker-list` (rowid-fenced pagination, fleet liveness, `attention`, literal `nextAction`).
- **Release is an explicit ownership table** (`decideWorkerTerminalRelease`): only an `owned` resource can be settled, the archive is mandatory where reachable, and an owner whose process is proven exited can always get out of `retained` via `archive_status: unavailable`. User-taken-over, external, and transferred panes stay retained.
- **Settled-worker resume fence** (folds in #17651): a settled dispatch whose pane is still open is fenced at settlement, on stop/abandon/exit, and at startup; lifted on release, retain, takeover, and pane reuse.
- **Liveness is `live` / `unverifiable` / `exited` only**, from execution-host evidence. Fleet projection reads the evidence clock, not the relay delivery clock. A host-certified exit outranks the worker's settled state. `unverifiable` never authorizes stop, abandon, retry, or release, in code or in the guide.
- **Federation:** structured reads negotiate by `method_not_found` so every shipped host keeps transcript-first output; exited remote workers are closed before being reported closed; epoch fencing holds across peer restart, downgrade, and pairing rotation; no per-second forced capability probe.
- **Schema v35:** repairs databases stamped v34 by the pre-fix branch (mailbox_handle default, index predicates), drops the write-only `lifecycle_transition_receipts` ledger and five never-read v31 identity columns.
- **Schema v36:** `dispatch:<id>` mailboxes get a real consumer generation on `dispatch_contexts` and `remote_dispatch_attachments`, bumped and fenced in the same transaction on every re-attach (manual inject, worker-start, federated attach). A stale worker whose Dispatch moved to another process now gets `consumer_fenced` instead of silently acking the new worker's Delivery. Run mailboxes already worked this way.
- **Schema v37:** `dispatch_contexts` records its creator (`creator_handle`, `creator_pane_key`), so a coordinator's context-only self-dispatch is bookkeeping rather than a nesting parent; before this, one self-dispatch made every later `worker-start` from that coordinator fail the depth cap. Pre-v37 rows keep counting (fails closed).
- **Dispatch-mailbox ownership is checked, not inferred.** A `check` from a process whose pane no longer holds the Dispatch, or whose last Attempt was abandoned/failed and moved to another terminal, gets `consumer_fenced` instead of an empty inbox that reads as "no mail yet". `--peek`/`--all` stay readable. A paneless caller still gets `stable_pane_required` with the rebind recovery.
- **Liveness certification is stricter:** a `process_exited` stage whose termination reason is `unknown` (a stop that was issued but never observed) projects `unverifiable`, not `exited`. Federated `worker-show` carries the execution host's verdict and host kind instead of a local guess. A live, ready worker with nothing pending has `nextAction: none` rather than pointing at the `worker-show` that produced it.
- **Wire:** `workerShow` keeps `dispatch.task_id` next to `taskId` for shipped CLIs. `ask --json` uses the standard `{ok, result}` envelope like every sibling verb.
- **Migration start-version detection** treats the two v32 recovery columns as versioned. Before this, every shipped database stamped below 32 resolved to the v6 floor and replayed the whole chain (the v23 backfill synthesized 68 phantom retained workers on a real v30 profile). Verified on a copy of a real 62 MB v30 profile: starts at 30, no row delta, integrity ok, 11 ms.
- **Skill guide** rewritten as a ≤200-line kernel plus seven references, to the outcome-first standard (Result / Done / Safe failure first, conditions not case lists, one done bar, references loaded at the point of use). The canonical loop uses `worker-start --spec`, names `worker-list` for completion accounting, documents `--retry-request` / `request-show` / `--wait-submit`, and requires positive evidence before any stall action. The other seven guides get the same treatment in #18724, split out so this PR stays orchestration-only.
- **`rpc/methods/orchestration-*`** (126 flat files) regrouped into `orchestration/{worker,federation,messaging,runs,gates}/`.
## Why
User reports showed the same boundary failures: false `agent_prompt_stalled` causing duplicate sends (#15180), coordinators unable to trust screen scrapes, cold-parked terminals receiving a pointer without the submit, settled workers accumulating as live tabs and auto-resuming after restart, and no way to tell a stalled worker from a working one.
## Linked issues
Fixes #15180. Fixes #17935 (orchestration skill description is 866 characters; a guard now caps every bundled skill at 1,024). Supersedes #17651 (fence folded in). Advances #16660, #16522, #14907, #13047.
## Review record
This PR was reviewed adversarially after revival: eight independent lenses (lifecycle, mailbox, send, worker, federation, transcript, complexity, live ergonomics), each required to prove findings with a failing test. That produced 16 proven blockers, all fixed with red-then-green regression tests, followed by two re-review rounds and a third fix wave that caught 3 regressions introduced by the fixes and 7 fixes that missed their target; all closed. A final pass (five lenses incl. a live built-runtime smoke, then a re-review of the fix wave) found and fixed seven more, chiefly the stale-worker mailbox steal, the self-dispatch depth wedge, and the unproven-exit certification. Three independent Codex (gpt-6-astra) passes followed: the first found nothing new, the second found and fixed 3 defects (task-status reachability, WSL-local host classification, peer-capability epoch), the third found and fixed 6 (production PTY controller never installed settled writes, ambiguous in-flight pointer failures allowed duplicate replay, SSH/relay deadlines cut off a valid `--wait-submit`, stop-vs-exit race during inspection, and two release-recovery paths for vanished or exited terminals). The full record (findings, proof tests, triage, declines with reasons) is archived outside the repo.
**Rework after the live smoke.** A first live cross-host run on the shipped adhoc build (this Mac, a paired Windows host on the same build, a paired Mac on 1.4.195, and an SSH host) found a P1: a running local worker read `unverifiable`/`missing_status` because the fleet snapshot rows lacked the terminal handle the matcher keyed on. A 59-row failure table over every bug fixed during review showed the same two classes recurring: a fact dropped in transit through optional fields, and two authorities for one fact. Two blind designs (Opus, Codex) converged on the same mechanisms, and the scoped tranches landed here with red-then-green seam tests from the real producer to the real consumer, faults injected only at the transport or hook-ingest boundary:
- **Settlement (data-loss class):** one three-valued `WriteSettlement` (`accepted | refused{reason} | unverifiable{reason, bytesHandedToTransport}`) from the SSH multiplexer through daemon client, providers, controller, to pointer staging. No boolean, no rejection-as-third-state. The two silent degrades that fabricated a handoff are deleted; a provider that cannot settle refuses before any effect. Pointer text and Enter share the contract; a partial flush is `unverifiable`, never `refused`.
- **Evidence identity (false-liveness class):** fleet agent-status evidence is a tagged union (`binding: worker | pane | unresolved{reason}`, `clock: observed | delivery`) minted once at ingest, so a hook row captured on one process incarnation can never bind to a later dispatch on the same pane. The matcher's `!worker.paneKey ||` defaults are gone. One host-scope parser replaces two.
- **Small pre-merge items:** `capability_unsupported` from an old peer is no longer relabelled `host_unavailable`; a producer census test asserts every agent-status consumer path projects a pane-only hook row as `live`.
Two ergonomics defects the second live run surfaced on a real database are fixed here too: a pre-v3 dispatch already marked `completed` projected as `outcome_unknown` / `requiresAction: true` forever (three copies of the outcome ladder disagreed on legacy rows; now one resolver, legacy `completed` reads `succeeded` with nothing to act on, legacy `failed` stays actionable on the failure), and an unscoped `worker-list` enumerated the entire database (now defaults to the Run bound to the calling terminal, `--run` overrides, and the receipt's additive `scope` field says which).
A third live round on the shipped adhoc build of `b082443e1f` (same four hosts) plus an unscripted run in the user's own prompt style (a plain Claude Code shell, `/orchestration`, three workers, zero errors, bound-Run default confirmed) found two more branch defects, fixed with red-then-green tests: a worker freshly started on a paired server projected `unverifiable`/`host_indeterminate` with `requiresAction` for ~3 minutes, including after its own `worker_done`, because the host's federation observation returned `missing_liveness_verdict` for any PTY the liveness register had not yet swept (the host now reads a connected pane it owns locally as `live`; disconnected or SSH-scoped panes stay `unverifiable`); and six pre-v3 completed rows still carried an `input` category because settling through the task-status path or `failDispatch` never closed the Dispatch's pending question threads (both paths close them now, and schema v38 closes threads already pending on settled rows). The guide's `worker-start` examples now show `--model sonnet`, since an omitted model inherits the launcher's default.
A Codex adversarial pass on the tranche diff found one real design hole (identity minted at read time instead of ingest, now closed) and two daemon settlement paths that threw instead of settling (fixed). Two `@ts-nocheck` runtime mixins on these paths were extracted into checked modules; the repo-wide `@ts-nocheck` count is unchanged at 171.
Deletions during review: ~1,900 lines (write-only ledger, unread columns, dead v1 archive path, test harnesses shipped in prod, duplicated liveness and state-machine copies, self-capability checks that were compile-time true).
## Testing
- `pnpm typecheck:tsc:node|cli|web` clean
- `pnpm run check:code-quality:changed` 0 findings; `check:react-doctor:changed` 0
- `pnpm verify:bundled-skill-guides`, `verify:skill-bundle-manifest`
- full `pnpm test` on the integrated head: 72,332 pass / 292 skipped; the only failures were three non-PR files (two zsh live-shell suites hit a node-pty spawn-helper ENOENT while a concurrent native rebuild ran, 44/44 in isolation; `release-checkout.unit.test.ts` is a known 30 s load timeout that passes in isolation on `origin/main` too).
- CI on
|
||
|
Jinwoo Hong
|
b44ef1e59d |
fix(skills): narrow computer-use discovery boundary (#17736)
* fix(skills): narrow computer-use discovery boundary * chore: remove merge-formatting noise * fix(skills): name browser page automation surfaces |
||
 Brennan Benson
|
1a9e819c40 |
feat(skills): land remaining hybrid stubs (#9846)
* feat(skills): land remaining hybrid stubs * fix(build): exclude skill stub sources from packages |
||
|
Brennan Benson
|
f8b430f725 |
feat(skills): ship orca-cli as a first-generation hybrid stub (#9238)
* feat(skills): ship orca-cli as a first-generation hybrid stub Convert the installable orca-cli SKILL.md from a full fat guide into a hybrid discovery stub: a safe CLI resolver, an `orca skills get orca-cli` pointer, and a bounded read-only fallback for pre-guide binaries. The version-matched command reference now lives only in the Orca binary (embedded guide table, served by `orca skills get`), so the distributed file can no longer drift from the binary that runs the commands. - generator projects STUB_TOPICS from skill-stubs/<name>.md, reusing the guide's own frontmatter so the routing/description surface is unchanged; the embedded full guide (bundled-skill-guides.ts) is untouched. - manifest regenerated: orca-cli releaseRevision 32->33 as an append-only snapshot; existing fat installs classify `outdated` and get the targeted `npx skills update` nudge (no in-app writes). - tests: command-guidance assertions repointed to the guide source (their home now), plus stub-projection + safety coverage. Only orca-cli converts; the other skills stay fat. Per notes/skill-freshness-design.md, the E.3 pointer-compliance spike and the E.5 RC window remain before any further thinning. allowed-tools is intentionally not added yet (frontmatter kept byte-identical to the guide). * fix(skills): distinguish guide lookup failures * chore(skills): refresh released skill mapping |