v1.4.7

Timeout resilience: a session killed at the deadline now has its sealed review findings posted post-mortem instead of losing the whole pass. Review retry state is reset between attempts so a retried session can prepare and submit. Both skills now treat OPENCODE_DEADLINE_EPOCH as a hard kill and reserve the final ~10 minutes for sealing and submission.

v1.4.6

Autopilot passes now exit after pushing instead of sitting in gh pr checks --watch for a full CI cycle (the ~10+ min built-in floor on every fixing pass). Watching remains available when the request explicitly asks (watch / wait for CI / --watch).

v1.4.5

Bound gh-shim probe cost (DENIED negative cache, one probe round per run); scope comment-guard to this run's own comments by author+timestamp so concurrent runs don't fail each other; tighten review findings (~80 words) and require committable suggestion blocks when the fix is contiguous.

v1.4.4

Fix agent instructions that taught lowercase -f body=@file (posts the literal path string; caused the leaked @/tmp/opencode/rN.md replies and comment-guard failure on Trackdub#257). Examples and the autopilot reply guidance now use -F raw-field correctly.

v1.4.2

Document contents:write caller permission required by opencode-bot.yml and update caller example pins.

v1.4.1

Consolidates /oc fix and /oc autopilot into the single autopilot playbook: merge conflicts -> unresolved threads/feedback -> failing CI, with structured summary coverage. Scoped /oc fix requests still narrow the pass.

v1.4.0

Adds the bundled autopilot skill: one disciplined merge-readiness pass per run (merge conflicts -> unresolved review threads -> failing CI), available via /autopilot prompt or /oc autopilot comments. Includes a scoped context7 documentation rule for third-party API fixes; callers can pass CONTEXT7_API_KEY to enable it.

v1.3.2

Fix: skip review runs triggered by bot actors (bot-pushed synchronize events could never pass the agent's write-permission check).

v1.3.1

Fix: grant contents:write in the reusable bot workflow so use-github-token callers can push agent commits.

v1.3.0

What's changed

• Incremental re-reviews: after a submitted opencode-agent[bot]/github-actions[bot] review, the pinned diff covers only commits since that review's head (last-reviewed...head) instead of the full base...head range, so findings on unchanged code are not re-raised on every synchronize. First runs still review the full PR. The last-reviewed marker is read from the GitHub reviews API, so it cannot be spoofed and advances only on successful submission; pending reviews and unreadable or incomparable commits fall back to the full range.
• README: fixed stale provider list left over from the GitHub Models removal.

Full changelog: v1.2.0...v1.3.0