Repository navigation
Release Aiden Agent 0.59.0 - #557
Conversation
Bump package.json and package-lock.json so the Release desktop workflow publishes v0.59.0, and record Workspace Artifacts (#554) and the phone visual viewer host side (#514) as shipping in 0.59.0 in the plan index. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0182MtyqGMMUndt6G8h36npZ
The macOS "stalled setsid handshake" test flaked on main (run 39370017039): the fixture waited 1.5s for publication inside the shell test's 2s deadline, so a loaded runner that spent over 500ms starting the shell and fixture hit the deadline (timed_out) before the fixture failed closed (exit 75). Bound the wait at 1s and shorten the delayed-detach case to 200ms so every case keeps a full second of margin under the unchanged 2s deadline. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0182MtyqGMMUndt6G8h36npZ
Hermes Review BotConfidence: 5 Engine: SummaryPrepares the Aiden Agent 0.59.0 release and resolves a CI test flake in the macOS subagent shell runner suite. The release version is bumped across Confidence Score: 5/5Fully traced and verified all changed files, POSIX timing semantics in the C fixture, prebuilt manifest source hashes, and documentation contracts. 📁 Important Files Changed
FindingsNo findings. 📊 Sequence DiagramUnchanged or not applicable for this change. Machine-Readable Findings[]
|
![pullfrog[bot]](https://avatars.githubusercontent.com/in/1768019?s=60&v=4){kind=small}
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes across all four files, including release-version selection and the fixture's handshake, timeout, and cleanup behavior.
- Release metadata: Advances the desktop version to
0.59.0consistently in the manifest and lockfile; the declared version selectsv0.59.0for publication. - Fixture timing: Reduces the test-only publication wait to one second and successful delayed detachment to 200ms without changing the production runner or two-second command deadline.
- Plan status: Records Workspace Artifacts and desktop-host phone visuals as shipping in
0.59.0, retaining the native-build and Playwright follow-ups. - Validation: Six release-contract tests and the focused Linux shell-runner suite passed. Direct fixture checks confirmed delayed success and exit
75for both stall modes. The two Darwin-only integration cases and signed distribution were not locally verified; hosted CI was still running on this head.
gpt-6.1-sol ｜ 𝕏
The prebuilt manifests hash every .c file under native/subagent-shell-runner, including the test-only setsid fixture. The shipped runner is built from main.c alone, so its binaries are unchanged; refresh only the recorded source hash, computed with nativeHelperSourceHash. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0182MtyqGMMUndt6G8h36npZ
There was a problem hiding this comment.
✅ No new issues found.
Reviewed changes since the prior review at ac26fd82, focused on the prebuilt manifest correction in fe309bc6.
- Updated prebuilt source hashes for Darwin universal, Linux x64, and Linux arm64 to match the edited test fixture. Unchanged binary checksums are appropriate because production compilation excludes that fixture; the updated hashes restore agreement with the CLI's prebuilt-selection checks.
- Verified native manifest integrity with all six
scripts/native-helpers.test.mjstests passing, including source hashes, binary checksums, and executable bits across all three targets. The hostedverifycheck also passed on this exact head.
gpt-6.1-sol ｜ 𝕏
|
This PR doesn't touch iOS. It changes only the version, the plan index, a desktop test fixture in C and the prebuilt manifests. iOS ran only because the version bump selects every area. Every other check is green on this head, including the macOS runtime-subagents suite that this PR fixes. I haven't found a fix for this iOS failure yet, so I'm re-running the failed job once. If it fails again, I'll treat it as a real failure. Generated by Claude Code |
Requested by sbt · project thread
Before: main is at 0.58.0, and its latest CI run is red because one macOS subagent test flaked (
subagent-shell-runner-io.test.ts:414, "a stalled setsid handshake fails within the existing shell deadline", run 39370017039). The release job runs that same suite with no tolerance for flakes.After: the version is 0.59.0, and merging this publishes v0.59.0 with Workspace Artifacts (#554), the Tiles context panel (#555), the desktop side of full-screen phone visuals (#514, Remote revision 35) and the built-in browser Google sign-in fix (#556). The flaky test has twice the startup margin it had before.
How: the setsid test fixture waited 1.5s for publication inside the shell test's 2s deadline, so a loaded runner that spent more than 500ms starting the shell timed out before the fixture could fail closed with exit 75. The fixture now gives up after 1s, and the delayed-detach case waits 200ms instead of 1s, so every case keeps a full second of margin under the unchanged 2s deadline.
docs/plans/README.mdrecords what ships in 0.59.0.Validation:
subagent-shell-runner-io.test.tspasses 20/20 on Linux. The darwin-only stall cases run in this PR's macOS CI. Run by hand, the fixture exits 75 after about 1,005ms in both stall modes and exits 0 after about 206ms with delayed detach.Generated by Claude Code