AgentSkillsCN

09 Collab Coordinator

09 协作协调员

SKILL.md
--- frontmatter
tags: [skill, codex-exec]
skill_id: "09_collab_coordinator"

Skill 09 — Collab Coordinator

Purpose: Run Codex Collab as a Standard lane accelerator: coordinate read-only helpers and synthesize findings for the single writer.

Protocols:

  • [[docs/PROTOCOLS/LANES]]
  • [[docs/PROTOCOLS/COLLAB]]
  • [[docs/PROTOCOLS/VAULT_PROBE]]
  • [[docs/PROTOCOLS/EVIDENCE]]
  • [[docs/PROTOCOLS/RUN_LOGGING]]
  • [[docs/PROTOCOLS/CONTEXT_BUDGET]]

Inputs (from Orchestrator)

  • docs/tasks/<task_slug>.md (required)
  • Optional: specific questions / areas to investigate
  • Optional: helper count (default 2–4)

Outputs (must write)

  • docs/notes/<task_slug>_collab.md (brief + synthesis)
  • artifacts/collab_summary.json
  • [[docs/skills/09_collab_coordinator/SUMMARY]]

Run log (mandatory)

Create: artifacts/_runs/<timestamp>_09_collab_coordinator/


Steps

  1. Confirm the lane is Standard. If Slow triggers are present, stop and recommend upgrading to Slow lane.
  2. Produce a Collab brief that enforces:
    • single writer (Orchestrator)
    • helpers are read-only
    • evidence pointer format (path:Lx-Ly)
    • timebox and scope boundaries
  3. Define 2–4 helper roles (Context Scout / Risk Reviewer / AC Checker / Diff Reviewer).
  4. Collect helper outputs and synthesize:
    • key findings + evidence pointers
    • risks / open questions
    • recommended next skill(s)
  5. Write docs/notes/<task_slug>_collab.md.
  6. Write artifacts/collab_summary.json with:
    • task_slug
    • helpers (roles + assignments)
    • findings (bullets)
    • evidence (array of {path, from, to, why})
    • risks (array)
  7. Write SUMMARY.md.

Definition of Done

  • Collab note exists and declares single-writer rule
  • Collab summary JSON exists
  • Summary exists and links outputs