parallel-agents

name	parallel-agents
description	Multi-agent orchestration patterns. Use when multiple independent tasks can run with different domain expertise.
allowed-tools	Read, Glob, Grep

Parallel Agents

Coordinate, don't just broadcast.

Use Cases

• Good: Complex multi-domain tasks, comprehensive reviews (Sec + Perf + Test).
• Bad: Simple single-domain tasks.

Invocation Patterns

1. Single: "Use security-auditor to review auth".
2. Chain: "Explore -> Backend -> Test".
3. Context: "Frontend finding -> Test generation".

Available Agents

• orchestrator: Coordination.
• security-auditor: Security/Auth.
• penetration-tester: Red team/Exploit.
• backend-specialist: API/Node.
• frontend-specialist: React/UI.
• test-engineer: Testing/TDD.
• devops-engineer: CI/CD/Docker.
• database-architect: SQL/Schema.
• mobile-developer: RN/Flutter.
• debugger: Fixes.
• performance-optimizer: Speed.

---

🧠 Aletheia Reasoning Protocol (Orchestration)

1. Generator (Team Assembly)

• Roles: "Who is the 'Reviewer'? Who is the 'Doer'?".
• Dependencies: "Backend needs API spec first?".
• Synthesis: "How to merge conflicting advice?".

2. Verifier (Loop Check)

• Deadlock: "Agent A waiting for B, B waiting for A?".
• Noise: "50 pages for a 1-line fix?".
• Cost: "Worth the token cost?".

3. Reviser (Streamlining)

• Handoff: "Pass file path, not content.".
• Focus: "Tell Security Agent to ONLY look at auth.".

---

🛡️ Security & Safety Protocol (Multi-Agent)

1. Isolation: Context separation where possible.
2. Hallucination: Verify "Agent A said X" is true.
3. Command Authority: Only Orchestrator/User authorizes destruction.
4. Context Limit: Monitor token usage.

Synthesis Protocol

One unified report, not separate outputs.

1. Summary: What happened?
2. Contributions: "Sec found X, Backend found Y".
3. Actions: Critical fixes first.