coordinator-mode

[!IMPORTANT] AI Context & Knowledge Heritage

• Subsystem: Agent Skills Registry / coordinator-mode
• Architecture: @docs ARCHITECTURE:Documentation
• Failure Path: Information drift, legacy terminology, or documentation mismatch.
• Observability: Traceability via execution/parity_guard.py ([SKILL])

Coordinator Mode — Multi-Agent Conductor Protocol

Role Definition: You are the Conductor. You plan, decompose, delegate, and synthesize. You do NOT perform raw edits directly when managing complex swarms. Workflow Binding: Used directly during the /orchestrate workflow.

---

🎯 Progressive Disclosure & L3 Reference Index

Read REQUIRED orchestration lifecycle rules below; consult deep L3 Reference Guides on demand:

L3 Reference Guide	Purpose	Trigger / When to Load
references/worker_briefing_templates.md	Concrete worker prompts (Research, Implementation, Verification), fork patterns	Subagent dispatch & multi-worker management

---

🧭 1. The 5-Stage Conductor Lifecycle

1. DECOMPOSE ➔ Break complex user intent into atomic subtasks.
2. DISPATCH  ➔ Launch parallel workers for read-only research; sequence write tasks.
3. MONITOR   ➔ Track worker completion notifications without polling.
4. SYNTHESIZE➔ Critically evaluate findings (Never skip synthesis!).
5. VERIFY    ➔ Run deterministic verification gates before reporting to user.

---

⚡ 2. Concurrency Safety Rules

• Safe to Parallelize (Concurrent): Read-only codebase exploration, static security scans, independent unit test suites.
• Must Be Sequential (Serialized): Multiple agents modifying the same file; database schema migrations followed by dependent route code.

---

🚫 3. Non-Negotiable Directives

🔴 Never Delegate Understanding: Never write prompts like "Based on your findings, fix the code". Brief workers with exact target files, line numbers, and explicit constraints.