Subagent orchestration
Skill ArcanEdge-AI/codex-agent-playbook/skills/subagent-orchestration
Use when a coding task may benefit from Codex subagents for planning, engineering, review, testing, or documentation lookup. Helps the main agent delegate bounded work and verify outputs before accepting them.From its SKILL.md
npx -y skills add ArcanEdge-AI/codex-agent-playbook --skill subagent-orchestrationAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
2.0 KB, 387 tokens by cl100k_base, as published. Nobody here has run it
Subagent Orchestration Skill
The main agent is the senior developer and orchestrator. Subagents assist but do not own the outcome.
Use this skill when:
- the task is complex, multi-file, risky, or ambiguous
- a Planner could improve the work sequence or validation strategy
- an Engineer can safely handle a bounded implementation slice
- a Reviewer can inspect a diff or design for correctness and risk
- a Tester can analyze failing checks, logs, flakes, or validation gaps
- Docs can verify repository, framework, library, API, or platform behavior
Do not use this skill when:
- the task is trivial
- one coherent design judgment is required
- requirements are materially unclear
- subagents would edit the same files
- the main agent cannot verify the result
Workflow:
- Clarify the task goal and success criteria.
- Decide which work, if any, should be delegated.
- Choose from the Codex roles: Planner, Engineer, Reviewer, Tester, and Docs.
- Right-size the model when model selection is available:
- use cheaper or faster models for bounded, low-risk, easily verifiable work
- use stronger reasoning for planning, implementation strategy, meaningful review, ambiguous debugging, security-sensitive work, and high-impact changes
- Give each subagent a precise assignment:
- role
- goal
- context
- model / reasoning guidance
- scope
- non-goals
- permissions
- required evidence
- output format
- Wait for delegated results before accepting conclusions.
- Verify subagent claims against primary evidence.
- Inspect any changed files yourself.
- Accept, reject, or revise subagent recommendations.
- Report relevant subagent usage in the final response.
Never accept a subagent's conclusion solely because it sounds confident.
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.