Releases: gersmann/codex-review-action
Releases · gersmann/codex-review-action
v1.6.3
What's Changed
Full Changelog: v1...v1.6.3
v1.6.2
What's Changed
Full Changelog: v1...v1.6.2
OpenAI Codex Code Review + Code Act v1
Release Overview
- Version: v1 (initial public release)
- Type: GitHub Composite Action for autonomous PR reviews and optional comment-triggered edits
- Default model: gpt-5 (fast model: gpt-5-mini)
Highlights
- End-to-end PR review: summary + precise inline/file-level comments.
- Comment mode (“act”): reply to “@codex …” with focused edits, commit, and push to PR head.
- Deterministic PR-diff anchoring for inline comments via review comments API.
- Local context artifacts written to .codex-context/ for traceability.
Core Features
- Two modes:
- review: autonomous code review on PR events.
- act: respond to @codex instructions in PR comments (and review comments).
- PR-diff inline comments (no commit-diff mismatch).
- Robust JSON output parsing (handles ```json fenced blocks).
- Self-using workflow example under .github/workflows/codex-review.yml.
Developer Experience
- Standardized envs: CODEX_* (no legacy fallbacks).
- Defaults tuned for reliability:
- CODEX_MODEL=gpt-5
- CODEX_FAST_MODEL=gpt-5-mini
- CODEX_REASONING_EFFORT=medium
- Makefile targets via uvx: fmt (ruff format), lint (ruff --fix), type (mypy), qa.
- mypy configured to ignore missing stubs for third‑party libs (clean local typing for our code).
Anchoring & Context
- Inline anchors computed from PR unified diffs and posted with create_review_comment.
- Fallback to file-level comments when lines are outside diff hunks.
- Artifacts:
- ./.codex-context/combined_diffs.txt
- ./.codex-context/diffs/.patch
- ./.codex-context/pr.md (title, author, body, issue + review comments)
Comment-Triggered Edits
- Triggers: @codex …, @codex: …, @codex edit: …, /codex …
- Action posts a status reply (success/dry-run/no-changes/failure) to the PR conversation.
- Uses plan + apply_patch tools; pushes minimal diffs to the PR head (when not dry-run).
Debug & Logs
- Streaming:
- agent_reasoning_delta → raw text (no wrappers, no newlines).
- agent_message_delta → raw text (same behavior).
- Debug levels:
- DEBUG_CODEREVIEW=1 basic events + token streams
- DEBUG_CODEREVIEW=2 adds trace-level details (anchoring, HTTP events)
Configuration (Inputs)
- mode: review | act (default: review)
- openai_api_key: required
- model: default gpt-5
- reasoning_effort: minimal|low|medium|high (default: medium)
- fast_model, fast_reasoning_effort: used for dedup/fast passes in review mode
- debug_level, dry_run, stream_agent_messages
- Optional: codex_python_version, extra_pip_args
- Prompt selection (review mode): built-in review prompt is used by default
Workflow Example (Review)
- Triggers: PR opened/synchronize/reopened/ready_for_review
- Permissions: contents: read, pull-requests: write
- Uses: gersmann/codex-review-action@v1 with openai_api_key secret
Workflow Example (Act)
- Triggers: issue_comment.created, pull_request_review_comment.created with “@codex”
- Permissions: contents: write, pull-requests: write
- Uses: gersmann/codex-review-action@v1 with mode: act
Known Limitations
- Forked PRs may restrict pushes from base repo workflows; run in fork or adjust permissions.
- File-level fallbacks used when exact inline anchoring is not possible (outside-hunk lines).
v1.6.1
What's Changed
Full Changelog: v1.6.0...v1.6.1
v1.6.0
Improved review flow + JSON output validation
v1.5.0
v1.4.0
fix: additional_prompt in review action
Full Changelog: v1...v1.3.1
v1.3.0
v1.2.1
Deps Update