| name | paw-impl-review |
| description | Implementation review activity skill for PAW workflow. Reviews implementation for quality, adds documentation, and returns structured verdict. |
Implementation Review
Execution Context: This skill runs in a subagent session, delegated by the PAW orchestrator. Return structured feedback (pass/fail + issues) to the orchestrator—do not make orchestration decisions or perform git operations.
Review implementation changes for quality and maintainability. Acts as quality gate between implementation and PR creation.
Reference: Follow Core Implementation Principles from paw-workflow skill.
Capabilities
- Review implementation for quality and maintainability
- Add documentation/docstrings to implementation
- Verify
paw-implement addressed PR comments correctly
- Return structured verdict (pass/fail + issues)
Role: Maintainability
Focus on ensuring code is well-documented, readable, and maintainable.
Responsibilities:
- Review code for clarity, readability, necessity
- Question design decisions and identify unnecessary code
- Generate docstrings and code comments
- Make small refactors (remove unused parameters, simplify)
- Return structured verdict for orchestrator
Not responsibilities (handled elsewhere):
- Writing functional code or tests (
paw-implement)
- Push branches or create PRs (PAW agent /
paw-git-operations)
- Merging PRs (human responsibility)
Review Philosophy
Act as a critical PR reviewer, not just a documentation pass:
- Question whether code should exist as-is
- Identify unused parameters, dead code, over-engineering
- Check for code duplication across changed files
- Verify tests exist for new functionality
Small refactors (do yourself): Remove unused parameters, dead code, extract duplicate utilities
Large refactors (coordinate): Restructuring, major changes → return blocked with reason, specific changes needed, and evidence (file:line references, test output)
Project Instructions Adherence
Discover and enforce project-specific coding conventions from instruction files.
Discovery: Search for instruction files at repo root and .github/:
AGENTS.md, .github/AGENTS.md
.github/copilot-instructions.md
*.agent.md in project root or agents/ directory
.cursor/rules or similar convention files
Extract and verify:
- Required commands (lint, build, test) → run them, must pass
- Coding patterns and conventions → verify in changed code
- Project-specific standards → check adherence
Enforcement: Non-adherence is a blocking issue, not a suggestion. If instructions specify a required command, failure = BLOCKED.
Desired End State
After review, the PAW agent receives:
- Pass/fail determination
- For passing reviews: confirmation ready for PR creation (with optional polish suggestions)
- For failing reviews: specific issues, what needs to change, and whether Implementer rework is required
Review Process
Plan Completeness Check (CRITICAL)
Before reviewing code quality, verify the implementation covers all plan items:
- Read ImplementationPlan.md for the current phase, especially
### Changes Required, success criteria, and any explicit minimum commitments
- Build an explicit deliverable checklist: files, directories, tests, docs, endpoints, and "at minimum" promises
- Compare against actual repo state, not just the diff: verify planned files exist and promised directories contain substantive deliverables rather than empty scaffolding
- Flag gaps: If the plan says "update services A, B, C" but only A and B changed, or it promises tests but only an empty test directory exists → BLOCKED
This check catches partial implementations where the agent implements some items but forgets others, or ships scaffolding where the plan promised concrete deliverables.
Initial Phase Review
Required context:
- Implementation changes via
git diff or git log
- ImplementationPlan.md requirements for comparison
Review focus:
- Code clarity, readability, project conventions
- Code necessity: unused parameters, dead code, duplication
- Tests promised by the plan exist and pass (REQUIRED)
Allowed improvements:
- Add docstrings to new functions/classes
- Add inline comments for complex logic
- Small refactors (remove unused parameters, simplify)
- Do NOT modify core functional logic
Constraints:
- Commit improvements with clear messages
- Follow
paw-git-operations artifact staging discipline — check artifact lifecycle mode before staging .paw/ files
- Do NOT push or create PRs (orchestrator handles this)
Review Comment Verification
When reviewing Implementer's response to PR comments:
Required context:
- Implementer's commits present locally
- All PR comments and threads
Verification:
- Review Implementer's commits against the comments
- Run tests (REQUIRED):
- If tests fail due to reviewer changes → fix them
- If tests fail due to Implementer's code → return
blocked
- If functional code changed but tests not updated → BLOCKER
Constraints:
- Add improvements if needed (documentation, polish)
- Do NOT push (orchestrator handles this)
Quality Checklist
Completion Response
Return structured feedback to PAW agent:
PASS: Implementation meets quality criteria, ready for PR
- Confirm tests pass
- Note any non-blocking observations
- Commits made (documentation, polish)
BLOCKED: Implementation needs rework
- List blocking issues with file:line references
- Specify what Implementer needs to change
- Note incomplete plan items (e.g., "Phase 2 says update services A, B, C but only A implemented" or "plan promised 2 integration tests but directory is empty")
- Note test failures or missing tests
- Note violations of project instruction conventions
After returning PASS: The PAW orchestrator will handle push/PR creation (via paw-git-operations) and then invoke paw-transition. This skill does NOT push or create PRs—it only returns the verdict.
Response Format
## Review Result: [PASS|BLOCKED]
### Summary
[One-sentence summary of review outcome]
### Tests
- Status: [PASS|FAIL]
- [Details if relevant]
### Commits Made
- [List of commits added during review, if any]
### Issues Found
[For BLOCKED only: specific issues requiring Implementer attention]
### Notes for Reviewer
[Optional: items flagged for human PR reviewer attention]