dmitrl-dev/aisecurity/qa-fix-loop

┌─────────────┐     issues     ┌─────────────┐
│  Reviewer   │ ─────────────→ │   Fixer     │
│   Agent     │                │   Agent     │
└─────────────┘                └─────────────┘
       ↑                              │
       │         fixed code           │
       └──────────────────────────────┘

{
  "status": "NEEDS_FIX | APPROVED",
  "issues": [
    {
      "id": "QA-001",
      "severity": "HIGH | MEDIUM | LOW",
      "file": "path/to/file.py",
      "line": 42,
      "type": "SPEC_VIOLATION | CODE_QUALITY | SECURITY",
      "description": "Что не так",
      "suggestion": "Как исправить"
    }
  ],
  "approved_aspects": ["tests", "documentation"]
}

digraph qa_fix_loop {
    rankdir=LR;
    node [shape=box, style=rounded];
    
    code [label="New Code"];
    review [label="Reviewer Agent"];
    decision [shape=diamond, label="Issues?"];
    fix [label="Fixer Agent"];
    done [label="✅ Merged", style=filled, fillcolor=lightgreen];
    
    code -> review;
    review -> decision;
    decision -> fix [label="Yes"];
    decision -> done [label="No"];
    fix -> review [label="re-review"];
}

if iteration > MAX_ITERATIONS:
    escalate_to_human(
        reason="QA loop exceeded max iterations",
        issues=unresolved_issues
    )

1. Engine code → Reviewer Agent
2. Reviewer checks:
   - All payloads detected (Strike integration)
   - Interface compliance
   - Performance bounds
3. Issues → Fixer Agent
4. Loop until approved

1. C code → Reviewer Agent (специализированный)
2. Reviewer checks:
   - Memory safety
   - Protocol compliance
   - Cross-platform compatibility
3. Issues → Fixer Agent
4. Loop until approved