mirror of
https://github.com/stablyai/orca.git
synced 2026-09-22 00:02:31 +00:00
Jinwoo Hong
7d98c8e2f3
* test(mobile): record main's source-control RPC behaviour before migrating it 45 scenarios over 11 source-control senders, recorded from main so the step-4 migration has a frozen answer to compare against. Adapters mount the real exported senders as plain functions, so no React host or device is needed. The 73 existing goldens change header-only (`baseline`, `recorderSha256`): any new scenario re-digests the recorder, and the pinned baseline had drifted from main in `src/shared` so recording required bumping it. Content is byte-identical on all 73 — verified field-by-field against HEAD. Scenarios deliberately pin the empty-message cases (`sc-*-refused-empty-message`, `sc-*-rejected-empty-message`), because a refusal with no message falls back to the screen's copy while a transport error with no message does not, and the two paths are easy to collapse when a call site moves behind an acceptance policy. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * refactor(mobile): send the source-control domain through typed RpcOperations 13 of the domain's 14 files now send through a declared operation instead of the raw request port: 44 references to 0. The holdout is use-mobile-git-requests.ts, whose single reference is a `(method: string, params)` dispatcher that five other hooks feed `{ method, params }` action steps at runtime; typing it is a step model change, not a call-site move, so its line stays at 1. Fifteen operations over fourteen methods. Two of them read git.status, and that is deliberate: the Changes screen publishes the host payload verbatim while hosted-review preparation reads the normalized projection, which returns null when `entries` is not an array and drops entries missing a path. Sharing the projecting reader would change what the Changes list renders, so both are named. Four loads still read the refusal envelope before interpreting, through readMobileGitRefusal: two degrade to a capability-missing screen, one retries a selector that is not visible yet, and one falls back from files.openDiff to files.open. `isMobileGitUnavailable` consults the code *and* the message and no acceptance policy carries either through, so the alternative was parsing a code out of a message. No new acceptance policy was added. Every migrated site keeps two error paths where it had two: a refusal with no message falls back to the screen's copy, a transport rejection surfaces its own message verbatim and keeps its delivery-unknown mark. Collapsing them into one catch is what would have turned an unknown mutation into a failed one. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): re-digest the goldens after a lint fix in the new adapter recorderSha256 only, all 118 files; every recorded observation is byte-identical. Re-recorded fromc57de48fd0in a separate worktree so the goldens stay attributable to pre-migration product source — recording from this branch would have made the parity claim circular. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): drive the reply matrix over every scripted reply, and fail closed The matrix picked its driven request from a hardcoded prefix list and `continue`d past any family the list did not name. That was 10 of 23 families — every one the source-control migration added — with no red test to say so, which is why that migration's mutation evidence came down to single hand-written scenarios. `replyMatrixSites` now takes every completion step in a family's base scenario: 61 sites instead of 13, one golden per site, no judgement about which request is the "real" one and nothing to edit when a domain is added. A family that scripts no reply throws, a repeated request name throws, and a census test asserts every family in the manifest has a matrix. A variant's downstream replies are marked `optional` and answered only if the request is outstanding, so a diverged reply that ends the chain records the truth instead of failing on an unsent request. The `normal` partition replays the first fulfilled reply the family records for that request, rather than a payload the test file invented per family. Absent and null do not count — each is already a partition — so four sites with no other recorded success are inventoried in REPLY_MATRIX_NORMAL_RESULT_INVENTORY with a reason each, and an entry whose family later records a success fails. Two partitions added: a refusal and a transport rejection with no message. That is the axis that separates a refusal falling back to the screen's copy from a transport drop surfacing its empty message verbatim; without it the two paths produce the same text and collapsing them is invisible. Every source-control family carried a hand-written `*-empty-message` scenario for exactly that. 13 hand-written scenarios the matrix now covers are deleted: 8 `*-empty-message` cases plus sc-history-rejected, sc-commit-message-null-result, sc-eligibility- refused, sc-create-stops-on-push-refusal and sc-base-ref-rejected. Kept, with reasons, are the ones the matrix cannot reach: a different action or action args (sc-review-commit-*, sc-prefill-*, sc-create-{refused,rejected}-empty-message, sc-prerequisite-{publish,force-with-lease,skipped}), a payload shape rather than an envelope shape (sc-review-status-entries-not-array, sc-create-existing-review), and multi-request combinations (sc-base-ref-{unavailable,repo-fallback}, sc-reveal-timeout). Goldens: 118 -> 153. All 92 survivors changed by their `recorderSha256` line only; no recorded observation moved. Re-recorded from the pinned baseline in a separate tree so the goldens stay attributable to pre-migration product source. `matrix-hostedreview.create-intent-git.commit-1` fails on this branch, and it is a true positive: `hostReplyErrorTextOrFallback` stringifies a non-string in-band host error where main returned `result?.error || fallback` and passed the object through. Left failing — the fix is a product change, documented in the README. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * fix(mobile): keep main's in-band commit error pass-through The expanded reply matrix caught a real divergence the nine original partitions missed. Main returned `result?.error || 'Commit failed'`, passing a truthy non-string straight through under a `string` annotation; the migrated helper stringified it to "[object Object]". Stringifying is arguably better — downstream does `result.error.replace(...)`, which throws on an object and merely looks ugly on a string. But this migration's contract is that no behaviour changes, and shipping an unannounced improvement inside a refactor is exactly what the parity evidence exists to prevent. Restores the pass-through; the latent throw is its own ticket. No host sends this today (`git.commit` is typed `{success, error?: string}`), but nothing validates it and mixed client/host versions are normal. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): re-digest the goldens for the merged recorder Main inverted two guards in family-recordings.test.ts and pilot-recordings.test.ts. No behaviour change, but both files are inside recorderSha256, so all 153 goldens failed the header check after the merge. Re-recorded from16d1ab81d3in a separate worktree carrying main's product source and this branch's merged recorder, so the goldens still capture main's behaviour rather than the migration's. `baseline` moves from7ce8e18d07to16d1ab81d3because main touched src/shared/skills*.ts, which the record guard compares; that change moved no recorded observation. Every field except `baseline` and `recorderSha256` is byte-identical across all 153 files. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): intern each observation entry instead of the whole field Making the reply matrix fail closed took the goldens from 118 files / 1.53 MB to 153 / 5.35 MB, because a per-site golden replays the chain across 11 reply partitions and every checkpoint's sender, payloads, settlements and effects re-state the whole history that came before them. Format version 2 pooled those fields whole, so the shared prefix was stored once per checkpoint, and once per partition again. Version 3 pools each entry of a list or map field instead. `golden-value-pool.ts` declares the container per field rather than sniffing it from the value, so a projection that changes one fails loudly instead of silently switching encodings. 153 files / 5.35 MB becomes 153 / 2.78 MB; the family that drove this, hostedReview.create-intent, 2.0 MB over 12 sites becomes 792 KB. This is a re-encoding, not a re-observation. Every one of the 153 goldens resolves to the recording its version 2 file resolved to, checked field by field, and every header field except recorderSha256 and goldenFormatVersion is byte-identical. The three mutations this branch's coverage rests on fail exactly as before: the gitStatusProjectionRead acceptance policy 16 (13 matrix, 3 hand-written), interpret inside the request chain 5 (all matrix), and the rewrapped transport rejection 4 (all matrix). It also makes diffs smaller, which is the opposite of what version 2's note predicted when it rejected this. Adding a timeoutMs to the first git.status of the create-intent chain touches the same 16 goldens either way, but version 2 moves 17,100 lines / 1.03 MB and version 3 moves 3,764 / 0.20 MB, because a changed entry no longer rewrites every field value containing it. `readGolden` now also refuses a pool entry that does not hash to its own key, and one no checkpoint reads. Content addressing is what makes an entry shared between checkpoints safe to share; an unreferenced entry would be content in the file that nothing compares. Recorded from16d1ab81d3with this branch's recorder laid over it, per the README's flow. The record fence is unchanged. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb * test(mobile): make the matrix census and inventory checks able to fail Three review findings, all in the recorder, none in product code. The family census pushed every family unconditionally, so it could never differ from the manifest keys; it now records a family only when a site generated a test, which is independent of replyMatrixSites throwing on an empty list. REPLY_MATRIX_NORMAL_RESULT_INVENTORY was only consulted for a live site, so a stale entry retired silently; a new assertion fails on any entry that names no live (family, request). Both verified by mutation: an empty site list and a renamed inventory request each fail the suite. The value pool resolves hashes with Object.hasOwn so a malformed golden cannot read an inherited key. Re-recorded from16d1ab81d3with this recorder laid over main's product source, per the README. All 153 goldens move on recorderSha256 only. Claude-Session: https://claude.ai/code/session_01JNnE9qzUZMMnqpZWCqM3nb
Orca
中文 · 日本語 · 한국어 · Español · Français · Português
The AI Orchestrator for 100x builders.
Run Codex, ClaudeCode, OpenCode or Pi side-by-side — each in its own worktree, tracked in one place.
Download Orca
Features
Mobile CompanionMonitor and steer your agents from your phone — get notified when an agent finishes and send follow-ups from anywhere. |
 |
Parallel WorktreesFan one prompt across five agents, each in its own isolated git worktree — compare the results and merge the winner. |
 |
Terminal SplitsGhostty-class terminals with WebGL rendering, infinite splits, and scrollback that survives restarts. |
 |
Design ModeClick any UI element in a real Chromium window to send its HTML, CSS, and a cropped screenshot straight into your agent's prompt. |
 |
GitHub & Linear, NativeBrowse PRs, issues, and project boards in-app — open a worktree from any task and review without a context switch. |
 |
SSH WorktreesRun agents on a beefy remote box with full file editing, git, and terminals — auto-reconnect and port forwarding included. |
 |
Annotate AI DiffsDrop comments on any diff line and ship them back to the agent — review, edit, and commit without leaving Orca. |
 |
Drag Files to AgentsVS Code's editor with autosave everywhere — drag files or images straight into an agent prompt. |
 |
Orca CLIAgents drive Orca too — script every workflow with |
 |
Also in the box:
- Quick open — Search across worktrees, files, agents, commands, and repo context without leaving your flow.
- Account switcher & usage tracking — See Claude and Codex usage and rate-limit resets, and hot-swap accounts without re-logging in.
- Rich repo previews — Preview Markdown, images, PDFs, and repo docs in the workspace.
- Computer Use — Let agents operate desktop apps and visible UI when a workflow needs real interaction.
- Notifications and unread state — Know when an agent finishes or needs attention, then mark threads unread to come back later.
- And many, many more — we ship daily, so this list is perpetually behind. The changelog is the real feature list.
Supported Agents
Works with any CLI agent — if it runs in a terminal, it runs in Orca.
Claude Code
Codex
Grok
Cursor
GitHub Copilot
OpenCode
MiMo Code
Amp
OpenClaude
Antigravity
Pi
oh-my-pi
Hermes Agent
Devin
Goose
Auggie
Autohand Code
Charm
Cline
Codebuff
Command Code
Continue
Droid
Kilocode
Kimi
Kiro
Mistral Vibe
Qwen Code
Rovo Dev
+ any CLI agent
Install
Desktop — macOS, Windows, Linux
- Download from onOrca.dev
- Or grab a build directly: macOS Apple Silicon · macOS Intel · Windows (.exe) · Linux AppImage · All builds
- Running
orca serveon a headless Linux server? See the headless Linux server guide.
Or via a package manager:
# macOS (Homebrew)
brew install --cask stablyai/orca/orca
# Arch Linux (AUR) — or stably-orca-git to build from source
yay -S stably-orca-bin
Mobile Companion — iOS, Android
Pair with your desktop app to monitor and steer your agents from your phone.
- iOS: Download on the App Store or join TestFlight
- Android: Download APK 0.0.48 · Install guide
Community & Support
-
Discord: Join the community on Discord.
-
Twitter / X: Follow @orca_build for updates and announcements.
-
WeChat: Scan to join the Orca community WeChat group 8. Group 8 may be full; if so, scan the Group 9 QR code instead.
-
Feedback & Ideas: We ship fast. Missing something? Request a new feature.
-
Privacy: See the privacy & telemetry docs for what anonymous usage data Orca collects and how to opt out.
-
Show Support: Star this repo to follow along with our daily ships.
Developing
Want to contribute or run locally? See our CONTRIBUTING.md guide.
The relay that pairs the mobile app with a desktop host is also in this repository under
cloud/, with a separate pnpm workspace and setup guide.
Signed Builds
Windows code signing sponored/provided by SignPath.io, certificate by SignPath Foundation.
License
Orca is free and open source under the MIT License.