A reusable Agent Skill for writing bounded, auditable role briefs.
This skill does not require spawning sub-agents. It gives an agent a strict brief format for exploration, implementation, review, verification, and handoff work.
The skill turns vague complex work into a task contract:
- role,
- goal,
- context,
- allowed actions,
- ownership,
- forbidden actions,
- output format,
- stop condition.
That structure makes delegated work easier to review and makes solo agent work less likely to drift.
Copy the skill folder into a compatible agent skills directory:
skills/multi-agent/
For Codex-style local installation, copy it into your local skills directory and restart the agent.
This skill is Markdown-only and has no OS-specific runtime dependency. It works on macOS, Linux, WSL2, and Windows as long as the host agent supports local skills.
Model guidance:
- A normal coding model is enough for simple brief drafting.
- Use a stronger reasoning model for security review, architecture exploration, long-horizon implementation, or multi-agent coordination.
- For the author's heavier workflows, the known-good profile is Codex-style agent mode with a GPT-5.5-class reasoning model and
highorxhighreasoning when available. - This skill does not require multimodal input unless the delegated task itself involves images, screenshots, or visual artifacts.
After copying skills/multi-agent/ into a local skills directory, restart the host agent and ask it to draft a bounded brief for a read-only exploration task. If the host environment includes the Codex skill validator, the folder should also pass quick_validate.py.
Use it for:
- read-only code exploration,
- scoped implementation work,
- security or correctness review,
- verification after a patch,
- handoff between sessions or agents.
Do not use it for simple one-step tasks.
skills/multi-agent/SKILL.md
AgentsMD Kit defines project-level rules and durable memory. Multi-Agent Skill defines task-level role briefs. They are intentionally separate.