Sub-Agent Council

The Council provides a structured, multi-perspective review to eliminate ambiguous assumptions before coding, and audits changes before task completion.

---

Mode 1: Review Prompt (Pre-Execution Gate)

Trigger before planning or coding begins to dissect instructions and surface hidden constraints.

3 Evaluation Lenses:

1. Clarifier Lens (Ambiguity & Assumptions):
   • Identify implicit or subjective requirements.
   • Formulate clear, binary acceptance criteria (pass/fail).
2. Skeptic Lens (Edge Cases & Blast Radius):
   • Evaluate boundary and failure scenarios: empty input, malformed data, race conditions.
   • Flag potential destructive side effects on external files or database schemas.
3. Minimalist Lens (YAGNI & Lean Engineering):
   • Eliminate unnecessary abstraction layers (no factory pattern for a single implementation).
   • Ensure the solution uses native platform features or stdlib first.

Output Template:

### [Council: Prompt Review]
- **Identified Critical Assumptions**: [List of assumptions]
- **Measurable Acceptance Criteria**:
  - [ ] Criterion 1
  - [ ] Criterion 2
- **Minimal Execution Path**: [Maximum 3-5 sequential steps]

---

Mode 2: Review Project (Post-Execution Audit)

Trigger at the end of each task or before marking work as complete.

3 Audit Lenses:

1. Correctness Lens (Verification & Invariants):
   • Verify every acceptance criterion is demonstrably met.
   • Require terminal execution evidence showing exit code 0.
2. Security Lens (Safety & Resilience):
   • Audit input sanitization at trust boundaries, path traversal risks, and credential exposure.
3. Architecture Lens (Simplicity & Diff Hygiene):
   • Eliminate dead code, scratch files, and unrequested refactoring.

Output Template:

### [Council: Project Review]
- **Verdict**: [APPROVED | REVISION_REQUIRED | BLOCKED]
- **Verification Evidence**: [Command line + Exit code 0 output]
- **Critical Findings**: [Bugs / vulnerabilities if any]
- **Required Cleanup**: [Lines/files to revert or remove]