Review+

review+ — Compliance Review

Wraps superpowers:requesting-code-review. Requires superpowers to be installed.

Procedure

Phase A: Gather Review Context

1. Load the plan from docs/plans/.

2. Get the list of files changed since the plan was created:

   git diff --name-only [plan-commit]..HEAD
   

3. Invoke the scout agent to get current codebase state for comparison:

   Agent(subagent_type="scout", prompt="Get current state of these files: [changed files list]. For each file, report: symbol count, key exports, test coverage status.")
   

Phase B: Spec Compliance Review

1. Dispatch the typed spec reviewer with the per-task payload (role content lives in the agent definition):

   Agent(subagent_type="spec-reviewer", prompt="Task requirements: [full task text from the plan]. Implementer's report: [what was claimed/committed for this task]. Verify the implementation matches the spec — nothing more, nothing less.")
   

   If the typed agent is unavailable, fall back to a general-purpose subagent using the superpowers spec-reviewer prompt template.

2. The spec reviewer verifies, for each task in the plan:

   • Check: was the task implemented? (file exists, code present)
   • Check: does the implementation match the plan's specification?
   • Check: are the specified tests present and passing?
   • Check: were any plan tasks skipped or significantly changed?

3. Check stale test status:

   • For each changed source file, verify a corresponding test file was also changed
   • Flag any source edits without test updates as stale test violations

4. Check enforcement compliance (see active enforcement rules from session-start output):

   • Active enforcement rules followed in all test and source files (real dependencies in stack/E2E tests; mocks appropriate in unit tests)
   • All claims of success are backed by command output
   • Full-loop assertions present where applicable
   • No conditional test assertions (if (condition) assert(...))
   • No empty test bodies
   • No .skip without documented reason

Phase C: Code Quality Review (delegate to superpowers:requesting-code-review)

1. Invoke superpowers:requesting-code-review for process. Where it instructs Task tool with general-purpose type, instead dispatch the typed reviewer with only the per-task payload:

   Agent(subagent_type="code-reviewer", prompt="DESCRIPTION: [what was built]. PLAN_OR_REQUIREMENTS: [plan file path or task text]. BASE_SHA: [starting commit]. HEAD_SHA: [ending commit].")
   

   If the typed agent is unavailable, fall back to a general-purpose subagent using the superpowers code-reviewer prompt template.

2. Check code quality:

   • Files are focused (one clear responsibility per file)
   • Interfaces are well-defined
   • No unnecessary abstractions
   • No speculative generalization
   • Positive framing in code comments and error messages

Phase D: Review Report

1. Produce a two-stage review report:

   ## Review Report
   
   ### Stage 1: Spec Compliance
   - [ ] All plan tasks implemented
   - [ ] Implementation matches plan specification
   - [ ] No stale test violations
   - [ ] Active enforcement rules followed (see session-start output)
   
   ### Stage 2: Code Quality
   - [ ] Files are focused and well-structured
   - [ ] Interfaces are clean
   - [ ] No unnecessary complexity
   - [ ] Positive framing throughout
   
   ### Verdict
   PASS / FAIL with specific items to address
   

2. If PASS: the implementation is complete. Proceed to integration. If FAIL: list specific items, return to tdd+ to fix, then re-run verify+ and review+.

Skill Chain

After review+ passes:

• The implementation is complete
• Proceed to merge/PR decision (superpowers:finishing-a-development-branch)
• If branch discipline is active, finish with a PR (gh pr create via superpowers:finishing-a-development-branch) rather than a local merge.

Completion

Report one of these states when the skill finishes:

• DONE — Review passed (PASS verdict). All spec compliance and code quality checks confirmed.
• DONE_WITH_CONCERNS — Review passed with minor items noted for follow-up.
• BLOCKED — Review failed (FAIL verdict). Items must be fixed before proceeding.
• NEEDS_CONTEXT — Need user input to resolve a review question or accept a deviation.