Safety Rules

参见 [shared/core/safety-rules.md](../shared/core/safety-rules.md) — 所有安全规则从共享层加载，避免跨技能重复维护。

Iteration Manager

Supervises iterative testing, verification, and improvement of code.

Quick Commands

Iteration Workflow

┌─────────────────────────────────────────────────────────────┐
│                    ITERATION CYCLE                          │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│  ┌──────────┐    ┌──────────┐    ┌──────────┐    ┌────────┐│
│  │  TEST    │───▶│ ANALYZE  │───▶│ SUGGEST  │───▶│ APPLY  ││
│  │          │    │          │    │          │    │        ││
│  └──────────┘    └──────────┘    └──────────┘    └────────┘│
│       │                                                   │ │
│       │              ┌──────────┐                         │ │
│       └──────────────│ VERIFY   │◀────────────────────────┘ │
│                      │          │                           │
│                      └──────────┘                           │
│                            │                                 │
│                      ┌─────▼─────┐                          │
│                      │ CONVERGE? │                          │
│                      └─────┬─────┘                          │
│                     YES    │    NO                           │
│                      ┌─────┴─────┐                          │
│                      ▼           ▼                          │
│                   [DONE]    [NEXT ITERATION]                │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Iteration Process

Step 1: Execute Tests

# Run all tests
pytest tests/ -v --tb=short

# Run with coverage
pytest tests/ --cov=package --cov-report=term-missing

# Run specific test categories
pytest tests/ -m "not slow"  # Skip slow tests

Step 2: Analyze Results

Collect and analyze:

• Test pass/fail rates
• Coverage percentages
• Error patterns
• Performance metrics

Step 3: Generate Suggestions

Based on analysis, suggest improvements:

• Bug fixes for failing tests
• Code coverage improvements
• Performance optimizations
• Code style improvements

Step 4: Apply Changes

Implement suggested improvements and verify.

Step 5: Verify & Iterate

Compare with previous iteration and decide to continue or stop.

Quality Metrics

Primary Metrics

Primary metrics track the most important quality indicators. Test pass rate has a target of 100% with 40% weight. Code coverage has a target above 80% with 30% weight. Lint score has a target of 0 errors with 15% weight. Type check has a target of 0 errors with 15% weight.

Secondary Metrics

Secondary metrics provide additional quality signals. Test duration target is under 60 seconds. Code complexity target is under 10. Documentation coverage target is above 50%.

Iteration Report Format

# Iteration Report - Run #N

## Summary

| Metric | Previous | Current | Change |
|--------|----------|---------|--------|
| Pass Rate | 85% | 92% | +7% |
| Coverage | 72% | 78% | +6% |
| Errors | 5 | 2 | -3 |

## Issues Found

### Critical
- [Issue 1]: Description

### Major
- [Issue 2]: Description

### Minor
- [Issue 3]: Description

## Improvements Applied

1. Fixed [bug] in module X
2. Added test for Y
3. Refactored Z

## Next Iteration Focus

- [ ] Address remaining coverage gaps
- [ ] Fix failing edge case tests
- [ ] Improve performance of slow tests

## Convergence Check

- Score: 92/100
- Trend: Improving
- Recommendation: Continue iteration

Convergence Detection

Stop iteration when:

1. All tests pass AND coverage > 80%
2. No new issues detected for 2 consecutive iterations
3. Improvement rate < 1% for 3 consecutive iterations
4. User-specified iteration count reached

Usage Examples

Basic Iteration

User: Run 5 iterations on the current project

Agent: Starting iteration cycle...

Iteration 1/5:
- Tests: 45/50 passed (90%)
- Coverage: 68%
- Errors: 3
- Applying fixes...

Iteration 2/5:
- Tests: 48/50 passed (96%)
- Coverage: 74%
- Errors: 1
- Applying fixes...

Iteration 3/5:
- Tests: 50/50 passed (100%)
- Coverage: 82%
- Errors: 0
- Convergence detected!

Final Report:
- Total iterations: 3
- Time: 5 minutes
- Improvements: 12 fixes applied

Focused Iteration

User: Iterate until test coverage reaches 90%

Agent: Starting coverage-focused iteration...

Iteration 1: Coverage 68% → 72%
Iteration 2: Coverage 72% → 78%
Iteration 3: Coverage 78% → 83%
Iteration 4: Coverage 83% → 88%
Iteration 5: Coverage 88% → 91%

Target reached! Coverage: 91%

Stop and Resume

User: /iterate stop

Agent: Iteration stopped at #3
- Current state saved
- Resume with /iterate resume

User: /iterate resume

Agent: Resuming from iteration #3...

Rules

• [rules/testing-protocol.md](rules/testing-protocol.md) - Testing execution protocol
• [rules/quality-metrics.md](rules/quality-metrics.md) - Quality scoring system
• [rules/iteration-workflow.md](rules/iteration-workflow.md) - Detailed workflow
• [rules/anti-aigc.md](rules/anti-aigc.md) - Iteration report anti-AIGC detection rules

Integration with Other Skills

• Use /paper new <topic> and /paper search <keywords> from academic-writer to document iteration results and quality metrics as structured papers. After generating an /iterate report, feed it to /paper for formal documentation.
• Use /人话 <text> from humanizer to humanize generated iteration reports, removing AI-typical patterns from test analysis summaries. Pipe /iterate report output through /人话 for natural-sounding documentation.
• Use /python-project test from python-project-developer to generate test suites before starting iteration. Scaffold tests with /python-project test, then run /iterate <n> to refine them against quality gates.

Best Practices

Five practices guide effective iteration. First, start with baseline by running initial tests to establish metrics. Second, focus on one area by prioritizing critical failures first. Third, track progress by comparing metrics across iterations. Fourth, know when to stop by avoiding over-optimization. Fifth, document changes by keeping track of what was improved.

Troubleshooting

Quality metrics not converging

• Symptom: /iterate status shows metrics oscillating without improvement trend
• Fix: Increase iteration count --convergence-window 5 for wider sample; check if test suite is non-deterministic; add --metric-weight to prioritize stable metrics

Test suite too slow for rapid iteration

• Symptom: Each iteration runs for hours, making iterative cycles impractical
• Fix: Run /iterate <n> --fast to execute only smoke tests per iteration; run full suite every Nth iteration with --full-suite-every 5

Iteration consumes too much budget without results

• Symptom: Many iterations run but improvement delta is below 1%
• Fix: Check if early stopping threshold is too low; use /iterate stop to manually terminate; review iteration strategy with --strategy review

Edge Cases

• Flaky tests: Non-deterministic test results cause false convergence — tag flaky tests with @flaky; exclude from quality metrics calculation
• Performance benchmarks: Benchmark tests need warm-up iterations — set --warmup 3 to exclude first 3 runs from metrics
• Cross-branch iteration: Testing changes across multiple git branches — use --branch <branch> to target specific branch
• Generated code validation: Iteratively improving AI-generated code — use --validate-output to add output correctness checks beyond unit tests
• Zero-change iterations: If change delta is literally zero, auto-detected as "stuck" and iteration is terminated

AIGC-Aware Output

Iteration reports must include specific metrics with before/after numbers, not vague "improvement observed". Every finding must be specific: "SQL injection in /auth/login line 47" not "security issues found". Fix suggestions must be actionable: "change X to Y" not "consider improving". See rules/anti-aigc.md for complete anti-AIGC detection rules.

Version History

See Also

• /python-project test from python-project-developer — Generate test suites for iteration
• /architect phase from master-architect — Quality gates during phased development
• /把关 check from ba-guan — Pre-publish quality validation