Code Reviewing

Perform AI-assisted code review immediately before escalating to human review. Ensures humans always see fresh, relevant code analysis when making decisions.

When to Use

•After completing implementation work (executing stage)
•Before any human checkpoint/escalation
•When /AI-review is invoked
•Before creating PRs or requesting human review

Quick Reference

Stage	Focus Areas
concept	Clarity, scope, testable criteria
planning	Design quality, pattern adherence
executing	Security, logic, architecture, tests
testing	Integration, coverage, regressions
releasing	All findings addressed, docs complete

Review Process

Phase 1: Gather Context

•Identify changed files - git diff or compare to last review
•Load task context - Read task description, acceptance criteria
•Check stage - Adjust review focus based on current stage

Phase 2: Run Automated Checks

bash

# Python projects
python -m flake8 src/ || echo "No flake8"
python -m mypy src/ || echo "No mypy"

# JavaScript/TypeScript projects
npm run lint || echo "No linter"
npm run typecheck || echo "No type checker"

Phase 3: AI Analysis

See references/review-checklist.md for complete checklist.

Key areas:

Category	Critical Checks
Security	Input validation, no hardcoded secrets, parameterized queries
AI-specific	No hallucinated APIs, follows project patterns, appropriate abstraction
Logic	Edge cases (null, empty, boundary), error handling, no infinite loops
Architecture	Follows conventions, correct dependency flow, no circular deps

Phase 4: Generate Report

markdown

## AI Code Review Report

**Task:** [ID] - [Title]
**Stage:** [current_stage]
**Files reviewed:** [count]
**Review date:** [timestamp]

### Summary
[1-2 sentence overall assessment]

### Findings

#### Critical (must fix)
- [ ] [Finding with file:line reference]

#### Warnings (should fix)
- [ ] [Finding with file:line reference]

#### Suggestions (consider)
- [ ] [Finding with file:line reference]

### Automated Check Results
- Linting: [pass/fail/skipped]
- Type check: [pass/fail/skipped]

### Recommended Actions
1. [Specific action item]

Phase 5: Save Review

•Update task notes - Append review summary
•Create artifact - Save to .claude/data/reviews/{task_id}_{timestamp}.md

Stage-Specific Focus

Concept Stage

•Concept clearly defined?
•Scope bounded and reasonable?
•Acceptance criteria verifiable?

Planning Stage

•Design follows established patterns?
•Avoids known anti-patterns?
•Plan is implementable?

Executing Stage (Full Review)

•All security checks
•All AI-specific checks
•All logic checks
•All architecture checks
•Tests adequate for changes?
•Code matches the plan?

Testing Stage

•Integration interfaces properly defined?
•Test coverage adequate?
•Regression test coverage exists?

Releasing Stage

•All previous findings addressed?
•Documentation complete?
•Release checklist items documented?

Common AI Code Mistakes

See references/common-ai-mistakes.md for complete patterns.

Issue	Detection
Hallucinated APIs	Verify imports/methods exist
Outdated patterns	Check for deprecation warnings
Missing edge cases	Test null, empty, boundaries
Security blind spots	Check input validation, query building
Over-abstraction	Is complexity justified?
Silent failures	Look for bare `except: pass`

Reference Files

File	Purpose
`references/best-practices-research.md`	Industry research on AI code review
`references/tool-capabilities.md`	What CodeRabbit, Copilot, Semgrep, SonarQube check
`references/review-checklist.md`	Complete checklist by category
`references/common-ai-mistakes.md`	Patterns specific to AI-generated code

Invocation

code

/AI-review --task-id=[ID] --stage=[current_stage]

Or simply /AI-review when context is clear from current work.