Release code-review-action 10.0.1

10.0.1 (2026-09-11)

Release Notes

Autopilot context loading is faster and more efficient in v10.0.1, with planning workflows now reusing fetched context instead of rebuilding it on every run.

✨ What's New

• Autopilot skips unused context work and loads optional skill references only when they are relevant to the current review, reducing overhead on every run.
• Standards summaries now report overflow explicitly instead of silently omitting applicable constraints, so reviewers see when limits are hit.
• Planning workflows accept explicit validated exploration briefs and reuse already-fetched Linear tickets, eliminating redundant fetches across Autopilot and Linear workflows.
• Exploration runs can refresh only affected sections, and stored Linear plans preserve reusable context evidence for subsequent runs.

Related issues

• #669: Reduce unnecessary context work and instruction overhead in Autopilot
• #671: Reuse planning context across Autopilot and Linear workflows

GitHub Issues

Issue	PR	Author
#671	#672	@awinogradov
#669	#670	@awinogradov

Performance

• autopilot: bound context and defer skill references (d71c112)
• autopilot: reuse planning and linear context (d1ed0ed)

Refactoring

• autopilot: share brief and input instructions (ed341f8)

Tests

• review: accept additional pr metadata fields (90ac035)

Release code-review-action 10.0.0

10.0.0 (2026-09-11)

Release Notes

10.0.0 (2026-09-11)

GitHub Issues

Issue	PR	Author
#664	#665	@awinogradov
#662	#663	@awinogradov

⚠ BREAKING CHANGES

• pr-monitor: pr-monitor requires a runtime that notifies the session
  when a background process exits; without one it reports the missing
  capability and stops instead of falling back to timed polling. It also
  adds a BLOCKED outcome and writes watcher state under the git common dir.

Claude-Session: https://claude.ai/code/session_01WM15kAVNHQTjxvke1Luypc
Entire-Checkpoint: 2fdf4671df31

• autopilot: --experts-review is no longer accepted and no flow launches an expert panel.

Features

• autopilot: remove the expert review panel (b53e55d)
• pr-monitor: wake only on actionable watcher events (b12493a)

Tests

• code-review-action: drop the expert panel removal sweep (9e04602)

GitHub Issues

Issue	PR	Author
#664	#665	@awinogradov
#662	#663	@awinogradov

⚠ BREAKING CHANGES

• pr-monitor: pr-monitor requires a runtime that notifies the session
  when a background process exits; without one it reports the missing
  capability and stops instead of falling back to timed polling. It also
  adds a BLOCKED outcome and writes watcher state under the git common dir.

Claude-Session: https://claude.ai/code/session_01WM15kAVNHQTjxvke1Luypc
Entire-Checkpoint: 2fdf4671df31

• autopilot: --experts-review is no longer accepted and no flow launches an expert panel.

Features

• autopilot: remove the expert review panel (b53e55d)
• pr-monitor: wake only on actionable watcher events (b12493a)

Tests

• code-review-action: drop the expert panel removal sweep (9e04602)

Release autopilot 9.0.1

9.0.1 (2026-09-11)

Release Notes

Autopilot's planning and execution workflows now reuse fetched context instead of rebuilding it on every invocation, cutting redundant token work across both Autopilot and Linear-backed flows.

✨ What's New

• Planning workflows now accept explicit validated exploration briefs and reuse already-fetched Linear tickets, so replanning a ticket no longer re-fetches context from scratch.
• Exploration can refresh only the affected sections of a plan rather than regenerating the full output, and stored Linear plans carry reusable context evidence forward to subsequent steps.
• Standards summaries now report overflow explicitly instead of silently dropping applicable constraints, making it visible when context limits caused a rule to be skipped.

Related issues

• #671: Reuse planning context across Autopilot and Linear workflows
• #669: Reduce unnecessary context work and instruction overhead in Autopilot

🐛 Bug Fixes

• Autopilot no longer loads optional skill instructions when the relevant skill is not active, eliminating unnecessary instruction overhead on every invocation.

Related issues

• #669: Reduce unnecessary context work and instruction overhead in Autopilot

GitHub Issues

Issue	PR	Author
#671	#672	@awinogradov
#669	#670	@awinogradov

Performance

• autopilot: bound context and defer skill references (d71c112)
• autopilot: reuse planning and linear context (d1ed0ed)

Documentation

• autopilot: correct linear-plan gate punctuation (0e7b645)

Refactoring

• autopilot: share brief and input instructions (ed341f8)

Tests

• autopilot: cover invalid issue lookup arguments (c9e6423)

Release autopilot 9.0.0

9.0.0 (2026-09-11)

Release Notes

9.0.0 (2026-09-11)

GitHub Issues

Issue	PR	Author
#664	#665	@awinogradov
#662	#663	@awinogradov

⚠ BREAKING CHANGES

• pr-monitor: pr-monitor requires a runtime that notifies the session
  when a background process exits; without one it reports the missing
  capability and stops instead of falling back to timed polling. It also
  adds a BLOCKED outcome and writes watcher state under the git common dir.

Claude-Session: https://claude.ai/code/session_01WM15kAVNHQTjxvke1Luypc
Entire-Checkpoint: 2fdf4671df31

• autopilot: --experts-review is no longer accepted and no flow launches an expert panel.

Features

• autopilot: add pr watcher that waits outside the model (cb0c214)
• autopilot: remove the expert review panel (b53e55d)
• pr-monitor: wake only on actionable watcher events (b12493a)

GitHub Issues

Issue	PR	Author
#664	#665	@awinogradov
#662	#663	@awinogradov

⚠ BREAKING CHANGES

• pr-monitor: pr-monitor requires a runtime that notifies the session
  when a background process exits; without one it reports the missing
  capability and stops instead of falling back to timed polling. It also
  adds a BLOCKED outcome and writes watcher state under the git common dir.

Claude-Session: https://claude.ai/code/session_01WM15kAVNHQTjxvke1Luypc
Entire-Checkpoint: 2fdf4671df31

• autopilot: --experts-review is no longer accepted and no flow launches an expert panel.

Features

• autopilot: add pr watcher that waits outside the model (cb0c214)
• autopilot: remove the expert review panel (b53e55d)
• pr-monitor: wake only on actionable watcher events (b12493a)

Release agents-skills-sync-action 2.0.0

2.0.0 (2026-09-11)

Release Notes

2.0.0 (2026-09-11)

GitHub Issues

Issue	PR	Author
#662	#663	@awinogradov

⚠ BREAKING CHANGES

• autopilot: --experts-review is no longer accepted and no flow launches an expert panel.

Features

• autopilot: remove the expert review panel (b53e55d)

GitHub Issues

Issue	PR	Author
#662	#663	@awinogradov

⚠ BREAKING CHANGES

• autopilot: --experts-review is no longer accepted and no flow launches an expert panel.

Features

• autopilot: remove the expert review panel (b53e55d)

Release code-review-action 9.0.0

9.0.0 (2026-09-07)

Release Notes

pr-monitor now exits at READY_FOR_REVIEW by default instead of blocking until a human approves, shortening autonomous delivery loops for CI pipelines that chain on monitor completion.

✨ What's New

• pr-monitor exits with Status: READY_FOR_REVIEW once checks have settled and all threads are answered, so autonomous delivery pipelines no longer stall waiting for a human approval gate — pass --wait-for-approval to restore the previous blocking behaviour.
• pr-monitor exits with Status: CHANGES_REQUESTED when a reviewer's verdict on the current head stands after every thread was answered without code changes, giving pipelines a clear signal to stop retrying.
• Later human feedback is handled by re-running /autopilot:pr-monitor or /autopilot:pr-resolve, keeping the handoff model explicit.

Related issues

• #656: Stop PR monitoring after autonomous delivery without waiting for human review

⚠️ Breaking Changes

pr-monitor no longer waits for human approval by default

run, run-primed, and linear-run all end at READY_FOR_REVIEW in v9.0.0; any workflow or script that relied on pr-monitor blocking until a human approves will now proceed immediately at that point instead.

Migration: pass --wait-for-approval to pr-monitor (and to the run, run-primed, and linear-run entry points that delegate to it) to keep the previous blocking behaviour.

GitHub Issues

Issue	PR	Author
#656	#657	@awinogradov

⚠ BREAKING CHANGES

• pr-monitor: pr-monitor no longer blocks until a human approves; run,
  run-primed, and linear-run end at READY_FOR_REVIEW. Pass --wait-for-approval
  to keep the previous behaviour.

Claude-Session: https://claude.ai/code/session_01BgoWVafn1Jxhdd2HdQBkJ9
Entire-Checkpoint: ce9f7c416776

Features

• pr-monitor: exit at ready-for-review hand-off (71ee4f7)

Tests

• code-review-action: guard delivery contract (db23754)

Release autopilot 8.0.0

8.0.0 (2026-09-07)

Release Notes

pr-monitor now exits at READY_FOR_REVIEW by default instead of blocking on human approval — pass --wait-for-approval to restore the previous behaviour.

✨ What's New

• run, run-primed, and linear-run complete at READY_FOR_REVIEW once checks settle and all threads are answered, so autonomous delivery pipelines no longer stall waiting for a human reviewer; re-run /autopilot:pr-monitor or /autopilot:pr-resolve to handle later feedback.

Related issues

• #656: Stop PR monitoring after autonomous delivery without waiting for human review

⚠️ Breaking Changes

pr-monitor exits at READY_FOR_REVIEW instead of waiting for approval

pr-monitor previously blocked until a human approved the PR; it now ends as soon as checks have settled and no review thread is unanswered, exiting with Status: READY_FOR_REVIEW. Any automation or workflow that relied on the agent completing only after approval will stop receiving that signal.

Migration steps:

1. Audit any scripts or CI jobs that treat a pr-monitor exit as "PR approved" — they must now check the status output explicitly.
2. Pass --wait-for-approval to /autopilot:pr-monitor (and to run, run-primed, or linear-run if you invoke them with monitoring) to restore the previous blocking behaviour.
3. To handle subsequent reviewer feedback, schedule a follow-up run of /autopilot:pr-monitor or /autopilot:pr-resolve after the PR receives human review.

GitHub Issues

Issue	PR	Author
#656	#657	@awinogradov

⚠ BREAKING CHANGES

• pr-monitor: pr-monitor no longer blocks until a human approves; run,
  run-primed, and linear-run end at READY_FOR_REVIEW. Pass --wait-for-approval
  to keep the previous behaviour.

Claude-Session: https://claude.ai/code/session_01BgoWVafn1Jxhdd2HdQBkJ9
Entire-Checkpoint: ce9f7c416776

Features

• pr-monitor: exit at ready-for-review hand-off (71ee4f7)

Documentation

• autopilot: describe ready-for-review outcome (2b8f952)
• run: align intro with ready-for-review hand-off (94a690d)

Release autopilot 7.5.2

7.5.2 (2026-09-07)

Release Notes

The pdf-create skill's renderer dependencies are updated, including @react-pdf/renderer 4.9.0 which adds PDF/A conformance support.

✨ What's New

• The pdf-create skill can now produce PDF/A-1/2/3 (b-level) output by passing a conformance prop to the Document component, enabling archival-grade PDFs with XMP metadata and sRGB OutputIntent — fonts must be registered (not the standard 14 built-ins) to fully validate.

CI

• deps: bump the pdf-create-renderer group (d29afc6)

Release release-action 1.5.0

1.5.0 (2026-09-02)

Release Notes

AI-generated release notes, PR descriptions, and review replies are now capped to one sentence per item across all autopilot outputs.

✨ What's New

• Release note bullets, PR descriptions, issue bodies, and review replies are each limited to one sentence — one fact and one reason — eliminating multi-paragraph prose from all autopilot outputs.

Related issues

• #643: Make generated prose short and confident across autopilot outputs

GitHub Issues

Issue	PR	Author
#643	#644	@awinogradov

Features

• autopilot: make generated prose short and confident (5e6895e)
• release-action: emit one-sentence release note bullets (ac51009)

Release code-review-action 8.1.4

8.1.4 (2026-09-02)

Release Notes

Autopilot skill loading is restructured so Claude reads large reference sections on demand instead of upfront, reducing session token usage on every PR review and react run.

✨ What's New

• Rarely-reached sections of preflight-check, commits-create, pr-update, and pr-monitor are now stored in per-skill references/ files and loaded only when needed, cutting the token overhead Claude incurs at the start of each autopilot session.

Related issues

• #647: Reduce autopilot session token usage from repeated skill body loads

GitHub Issues

Issue	PR	Author
#647	#648	@awinogradov

Refactoring

• autopilot: move unreached sections to references (2c8f445)

Tests

• code-review-action: guard skill body byte budgets (7ee5b2e)