Skip to main content
/review loads the code-reviewer agent to run a structured review of the implementation against the plan, requirements, and quality standards. It is the gate between /implement and /git — nothing ships without a review verdict.

​
What gets reviewed

The review covers four areas. Each one is checked independently and reported separately.

​
Plan compliance

  • Does the implementation match what was planned?
  • Are all acceptance criteria met?
  • Is there any scope creep — things added that weren’t planned?

​
Code quality

  • Is the code readable without needing comments to explain it?
  • Are functions doing one thing?
  • Is error handling present at system boundaries?
  • Are there tests for the new behavior?

​
Security

  • Is there any user input that isn’t validated?
  • Are there any secrets or credentials in the code?
  • Are there any new endpoints missing authorization?

​
Performance

  • Any N+1 queries — a loop containing a database call?
  • Any synchronous blocking I/O in an async context?
  • Any unbounded data fetching — missing LIMIT on queries?
  • Any unnecessary re-computation inside loops?

​
Issue severity

SeverityMeaning
CriticalBlocks shipping — security vulnerabilities, broken functionality
HighShould fix before shipping — significant quality issues
MediumFix soon — technical debt, missing tests
LowNice to have — style, minor improvements
Only Critical and High issues block the gate. Medium and Low issues are surfaced as suggestions and can be addressed after shipping.

​
Review output format

The code-reviewer agent produces a structured report: 
=== CODE REVIEW ===
Task reviewed: [task N from plan]
Reviewer: code-reviewer agent

PLAN COMPLIANCE: PASS
- All acceptance criteria met
- No scope creep detected

CODE QUALITY: 2 issues
- MEDIUM: auth.js:45 — function is 87 lines, extract token validation to separate function
- LOW: Missing error message context in catch block at user.js:102

SECURITY: PASS (no auth/input handling changes)

PERFORMANCE: PASS

OVERALL: APPROVED WITH SUGGESTIONS
Blocker count: 0 | Suggestions: 2

​
Verdicts and what they mean

No issues found. The implementation is ready to ship. The gate to /git is open.

​
Agents involved

AgentWhen invoked
code-reviewerPrimary — runs the full review checklist
security-engineerInvoked when the changes include auth, input handling, credentials, or new endpoints
The receiving-code-review skill is also loaded when applying feedback from a CHANGES REQUIRED verdict, to ensure feedback is incorporated correctly.
If the review returns CHANGES REQUIRED, SuperAntigravity loads the receiving-code-review skill to guide you through applying the feedback systematically. After fixes are applied, /review is run again before the gate opens.

Build docs developers (and LLMs) love

Get started for freeTalk to us