This release adds 3 notable features for engineering teams evaluating rollout.
✓ No known CVEs patched in this version
Topics
+6 more
Summary
AI summarySupervisor evaluations now automatically retry after stalls and PR comment commands limited to authors.
Full changelog
🚢 Patch release with reliability and review-quality improvements.
Highlights
- Model supervisor evaluations now retry after supervisor subprocess stalls, timeouts, or forced kills while preserving the existing executor attempt evidence.
galley daemon stop --forcenow cleans up known executor and supervisor child process groups before reporting success.- PR comment commands such as
/galleyare now accepted only from the pull request author, in addition to the existing GitHub trust checks. - Generated PR titles now use more of GitHub's 256-byte title limit before truncating.
- Claude and Codex supervisor prompts now treat task input files as implementation source materials and check that requested core mechanisms are preserved.
- The Claude executor prompt now uses ordered Step gates, source-material classification, quality-profile rule mapping, and a final self quality gate.
Notes
- No new CLI flags, task YAML fields, or profile fields were added for supervisor retry behavior.
- Task files created before this release do not have a persisted PR author, so PR comment commands are rejected until the author is recorded.
Weekly OSS security release digest.
The CVE patches and breaking changes that affected production tools this week. One email, every Sunday.
No spam, unsubscribe anytime.
Share this release
About Galley
Local-first runtime for supervised AI coding tasks: isolated git worktrees, supervisor review against acceptance criteria, retry/escalate loops, on-disk run evidence, and PR handoff. Supports Codex CLI and Claude Code. Go, MIT.
Related context
Beta — feedback welcome: [email protected]