Iterate on Feedback

Goal

Execute a reflection loop that implements fixes, validates changes, and iterates until code is merge-ready or stopping criteria are met.

Inputs

• Gap analysis report (from analyze-code-gaps skill)
• Maximum iterations (default: 5)
• Stopping criteria (default: all blockers resolved + tests passing)

Process

Iteration Loop

For iteration 1 to MAX_ITERATIONS:
  1. SELECT: Choose highest priority unresolved gap
  2. IMPLEMENT: Apply the fix
  3. VALIDATE: Run tests and checks
  4. REFLECT: Analyze results and update gap analysis
  5. CHECK: Evaluate stopping criteria
  6. REPORT: Update progress
  
  If stopping criteria met: EXIT SUCCESS
  If no progress in 2 iterations: ESCALATE
  If max iterations reached: EXIT WITH STATUS

1. Selection Phase

• Load current gap analysis
• Filter for unresolved items
• Sort by priority score (descending)
• Select top item or batch of independent items
• Show user what will be fixed: "Fixing Gap-1: Missing null validation"

2. Implementation Phase

• Read affected file(s)
• Apply recommended fix or generate solution
• Follow existing code style and patterns
• Add necessary imports/dependencies
• Update related documentation
• Log changes made

3. Validation Phase

Run comprehensive checks:

# Syntax check
npm run lint

# Type checking (if applicable)
npm run type-check

# Unit tests
npm test

# Integration tests (if quick)
npm run test:integration

# Security scan (if available)
npm audit

Capture results:

• Tests passing/failing
• New errors introduced
• Performance impact
• Coverage changes

4. Reflection Phase

Analyze validation results:

• Success: Mark gap as resolved, update score
• Partial Success: Mark as partially addressed, note remaining work
• Failure: Rollback changes, mark as needs different approach
• New Issues: Add to gap list, adjust priorities

Update state file: tasks/code-review-[pr-number]-state.json

{
  "iteration": 2,
  "gaps_resolved": ["Gap-1", "Gap-3"],
  "gaps_remaining": ["Gap-2"],
  "test_status": "passing",
  "merge_readiness": 75,
  "changes_made": [
    {
      "gap_id": "Gap-1",
      "file": "src/auth.js",
      "lines_changed": 3,
      "commit": "abc123"
    }
  ]
}

5. Stopping Criteria Check

Success Criteria (exit with success):

• All blocking gaps resolved
• All tests passing
• Merge readiness score >= 80
• No new issues introduced

Escalation Criteria (pause for human):

• No progress in last 2 iterations
• Same test keeps failing
• Architectural change needed
• Unclear how to proceed

Max Iterations (exit with report):

• Reached iteration limit
• Show final status and remaining work

6. Progress Reporting

After each iteration, report:

## Iteration 2 Complete ✓

**Changes Made**:
- ✅ Fixed Gap-1: Added null validation to auth.js
- ✅ Fixed Gap-4: Updated error messages

**Validation Results**:
- Tests: 127/127 passing ✓
- Linter: No errors ✓
- Type Check: Passed ✓

**Progress**:
- Gaps Resolved: 2 → 4 (+2)
- Merge Readiness: 45 → 75 (+30)
- Remaining Blockers: 3 → 1 (-2)

**Next Iteration**:
Will address Gap-2 (SQL injection fix)

Continue? (yes/no)

Commit Strategy

Per-Iteration Commits

After successful validation:

git add [affected-files]
git commit -m "fix(review): address Gap-1 - add null validation" \
           -m "- Added null check in validateUser function" \
           -m "- Added unit test for null input case" \
           -m "- Resolves review comment from @senior-dev" \
           -m "PR #123 iteration 2"

Commit Message Format

<type>(review): <brief description>

- Detailed change 1
- Detailed change 2
- Test additions/modifications

Resolves review comment from @reviewer
PR #[number] iteration [n]

Rollback Strategy

If validation fails:

1. Identify what broke
2. Revert changes: git checkout -- [files]
3. Analyze why it failed
4. Try alternative approach or escalate
5. Log failed attempt for learning

Interaction Protocol

• Before each iteration: Show what will be fixed, ask for approval
• After each iteration: Show results, ask to continue
• On failure: Explain what went wrong, propose next steps
• On success: Celebrate progress, show remaining work
• On escalation: Clearly state why human input needed

Safety Guardrails

• Never skip test validation
• Always commit working state before next iteration
• Preserve original code in comments if major refactor
• Don't modify files outside scope of review comments
• Don't introduce new dependencies without approval
• Maximum 5 iterations to prevent infinite loops

Output Artifacts

Iteration Log

Save to tasks/code-review-[pr-number]-iterations.md:

# Iteration Log - PR #123

## Iteration 1
**Started**: 2025-10-21 11:30:00  
**Duration**: 8 minutes

**Target**: Gap-1 (Missing null validation)
**Changes**: Added null check in auth.js:42
**Tests**: 127/127 passing
**Result**: ✅ Success

## Iteration 2
**Started**: 2025-10-21 11:38:00  
**Duration**: 15 minutes

**Target**: Gap-2 (SQL injection)
**Changes**: Converted to parameterized query in query.js:78
**Tests**: 127/127 passing
**Result**: ✅ Success

## Summary
- Total Iterations: 2
- Gaps Resolved: 2
- Final Merge Readiness: 85/100
- Status: Ready for re-review

References

• See reference.md for iteration examples
• See code-review-orchestrator agent for workflow context
• See analyze-code-gaps skill for gap prioritization