Menü
Automated CI code review for GitHub PRs using OpenAI Codex. Validates PR descriptions, checks implementation alignment, enforces Quest architecture boundaries, and maps test coverage to acceptance criteria when PRs move to ready-for-review.
| name | ci-code-reviewer |
| description | Automated CI code review for GitHub PRs using OpenAI Codex. Validates PR descriptions, checks implementation alignment, enforces Quest architecture boundaries, and maps test coverage to acceptance criteria when PRs move to ready-for-review. |
Automated code review for CI pipelines. Designed to run inside GitHub Actions
via openai/codex-action, producing a single structured PR comment.
This skill extends .skills/code-reviewer/SKILL.md with CI-specific
adaptations: no interactive prompts, PR description validation, manifest
pre-checks, and concise output formatting for automated comments.
Use this skill when:
Do NOT use for:
code-reviewer skill instead)plan-reviewer skill)Before starting the review, read:
AGENTS.md -- architecture boundaries and coding conventions.skills/code-reviewer/SKILL.md -- severity model and review baselineIf any file is missing, note it in the output and continue with available context.
Classify findings using these levels, and include the level in each finding:
Rule: avoid comment-only style polish and trivial corrections. Prefer high signal.
This step is mandatory and must appear first in the output.
Check the PR description for:
.quest/*/phase_01_plan/plan.md
or docs/implementation/....If neither condition is met:
If description exists but is very thin (fewer than ~20 words, little context):
Run ./scripts/validate-manifest.sh.
.quest-manifest
is updated in the diff.Verify that what the PR implements matches what it claims to do. Use the best available source of intent, in priority order:
Missing plan file is not a reason to skip alignment. Fall back to description.
Compare expectations vs diff:
For each acceptance criterion:
manual - not verifiedunclear - needs human reviewIf all criteria align, stay silent on this section.
git diff --name-only <base>...<head>.git diff --stat <base>...<head>..skills/, including .skills/quest/).ai/)scripts/).github/workflows/)docs/, AGENTS.md, quest markdown artifacts)Use these categories to decide which boundary checks and quality checks apply.
Apply Quest boundaries from AGENTS.md:
.skills/: skill definitions, review protocols, delegation docs.ai/: agent roles, templates, schemas, allowlist/context configscripts/: validation and utility shell scripts.github/: CI and repository automation config (especially workflow orchestration)docs/: documentation and guidance, no executable logicFor each changed file, verify:
Boundary violations are Must fix or Blocker depending on severity.
Review for:
Focus on behavior correctness, not style preferences.
PR-focused checks:
Secret leakage is always Blocker.
If acceptance criteria are available from Step 0.5:
For workflow/config/skill-only changes, acceptable evidence can include:
./scripts/validate-manifest.sh./scripts/validate-quest-config.shIf no criteria are available:
Review for clarity, maintainability, and consistency with existing patterns.
Markdown and YAML:
Shell scripts:
JSON/schema/config integrity:
.quest-manifest, skill index)Apply minimal diff principle: smallest correct change, no speculative refactors.
Keep output concise. Omit sections with no findings.
**PR Description:** PASS | FAIL - one-line explanation
**Summary:** 1-3 bullets. Recommendation: APPROVE / REQUEST CHANGES
**Findings** (only sections with issues):
**Blocker** - [path] Description and suggested fix
**Must fix** - [path] Description and suggested fix
**Should fix** - [path] Description and suggested fix
**Plan alignment** (only if gaps/scope creep found): list issues
**Test gaps** (only if gaps found): list missing coverage
**Architecture** (only if violations found): list violations
Line number guidance:
path:line when the line comes from the diff.path alone.Brevity rules:
Play Snake. Trigger for /play:snake, play snake, snake game, let's play snake, or requests to launch the Snake game.
Play a quest completion celebration animation. Use when the user invokes /celebrate, asks to celebrate a quest, or when a quest reaches the complete/archived state.
Review actual code implementations (Python, JavaScript, TypeScript, React) for correctness, maintainability, security hygiene, and adherence to project patterns. Use when reviewing pull requests to ensure code quality, proper error handling, adequate test coverage mapped to acceptance criteria, and alignment with architecture boundaries and review standards.
Generates commit messages by inspecting staged changes and repo history. Matches existing conventions (Conventional Commits or plain English), leads with intent, and appends the Quest co-author trailer. Use when the user asks for a commit message, help with git commit, or when reviewing staged changes for commit.
Implement an approved implementation plan step by step, producing small reviewable changes, mapping code/tests to acceptance criteria, and maintaining a lightweight decision log. Use when the plan/spec is already agreed and you want disciplined execution.
Create implementation plans with testable acceptance criteria, validation strategies, integration touchpoints, and risk analysis before coding begins.