Guide · Agent skill
How to use the Chief of Staff agent skill (chief-of-staff)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The chief-of-staff skill is an agent skill that turns your coding agent into a coordinator for a long-running goal. Instead of doing every task itself, the agent hands work to subagents, suggests recurring check-up schedules, and keeps a wiki of markdown pages that records decisions, user preferences and lessons. This guide explains what the skill does, how to install it, how to trigger it, and what to watch for before you use it.
Chief of Staffmattpocock/skills · MIT Download the skill Quick start
- 1Download the skill. Download the ZIP for chief-of-staff from the directory page and unzip it.
- 2Choose an install location. Place the skill folder in ~/.claude/skills/ for all projects, or in a project's .claude/skills/ folder for one project.
- 3Or upload it in the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the skill.
- 4State a long-running goal. Explain the goal, its expected duration, and that you want subagents and a wiki.
- 5Agree the wiki location. Confirm where the wiki folder should live before the agent writes its first page.
- 6Review schedules and changes. Accept or reject suggested check-up schedules and review environment changes as they are proposed.
What is an agent skill, and what does chief-of-staff make the agent do?
An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when it is relevant. The chief-of-staff skill is written for a session that will run for a long time toward one goal.
Once active, the skill asks the agent to think on two tracks at once. The tactical track asks how to complete the immediate task. The strategic track asks how to change the surrounding environment so the next task goes better.
- Delegates work to subagents, which are helpers that start with a fresh context window, so the main session's context stays small.
- Suggests recurring schedules that could help achieve the goal.
- Keeps a wiki: a folder of interlinked markdown pages with an index.md that outlives the session.
- Records decisions and user corrections as dated claims, and cites commits when a claim concerns code.
- Looks for environment improvements such as lint rules, coding standards files and better logs.
How do you install the chief-of-staff skill?
Download the ZIP for the skill from the directory page and unzip it. The skill folder contains agents/openai.yaml and SKILL.md. Place the unzipped skill folder in one of two locations: ~/.claude/skills/ to make it available across your projects, or a project's .claude/skills/ folder to limit it to that project.
If you use the Claude apps rather than a local tool, you can upload the skill instead. Open Settings, go to Capabilities, then Skills, and upload it there.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project folder
- Claude apps: Settings → Capabilities → Skills
How do you trigger the skill with natural requests?
The SKILL.md sets disable-model-invocation to true, which means the agent will not load this skill on its own from a casual request. You need to invoke it deliberately. Check your tool's documentation for how to call a skill explicitly, then state the long-running goal and ask for the coordination behaviour directly.
Good requests name the goal, its expected duration, and the fact that you want subagents and a wiki. The skill is aimed at a single session pursuing one goal, so say what that goal is.
- Use the chief-of-staff skill to lead a two-week migration of our auth module. Keep a wiki of decisions and suggest check-up schedules.
- Act as chief of staff for this goal. Delegate the research to subagents and record every decision we make in the wiki.
- Before the next step, read the wiki index and tell me what we already decided about the database layer.
What does a typical session look like?
At the start, the skill asks the agent to agree a location for the wiki with you before its first write. During the work, the agent starts subagents in the background so it can keep talking with you. Subagents communicate sparingly and pass context through pointers such as research notes, previous commits and wiki pages, rather than repeating information.
When you give a durable instruction or correction, the agent is asked to update every page that it touches and the index. Periodically it can suggest a lint schedule that looks for contradictions, orphan pages, missing links and claims that newer sources have overtaken.
- Ask the agent where the wiki should live. It must agree this with you before the first write.
- Expect the agent to delegate implementation and research rather than do it all itself.
- Accept or reject suggested schedules. The skill says to suggest them only where the harness permits.
What does the skill need, and what are the safety notes?
The skill lists no requirements, does not run scripts, and does not use the network. The SKILL.md itself contains no executable scripts. However, its instructions push the agent toward broad changes: it asks the agent to improve the environment proactively and to use every user message as a chance to search for improvements.
Because of this, supervise the session closely. The skill expects a persistent wiki folder and background agents, so check that your tool allows both before you start. Review any environment change the agent proposes or makes, especially to lint rules, logs and project conventions.
- No listed requirements. Runs scripts: no. Uses network: no.
- Expect a persistent wiki folder that outlives the session.
- Expect background subagents and suggested schedules.
- Review environment changes before accepting them.
Tips for getting the most out of it
Use the skill for a goal that really spans several days or sessions. For a quick one-off task, the overhead of subagents, a wiki and schedules is not worth it. Keep the wiki in a folder you can see and commit, so the pages persist.
Ask the agent to query the index before major decisions. The skill is built around reading index.md first, so this keeps the record useful.
- Choose a multi-session goal, not a small task.
- Keep the wiki in a visible, persistent folder.
- Ask for a query of index.md before each big decision.
- Ask for a lint pass periodically to catch orphan pages and stale claims.
What people use it for
- Running a multi-day engineering migration while the agent records decisions in a wiki.
- Keeping dated records of user preferences and corrections across several sessions.
- Delegating research to subagents while the main session stays focused on the goal.
- Suggesting a recurring lint schedule to find contradictions and orphan pages in a project wiki.
- Experimenting with multi-agent workflows in Claude Code or similar tools.
Frequently asked questions
Is the chief-of-staff agent skill free?
The skill is published under the MIT licence. The source does not list any price, so check the directory page for any further terms.
What do I need to run the chief-of-staff skill?
The skill lists no requirements. It does not run scripts and does not use the network. It does expect a persistent wiki folder and the ability to run background subagents.
Where do I install the chief-of-staff skill?
Put the unzipped skill folder in ~/.claude/skills/ for personal use, or in a project's .claude/skills/ folder. In the Claude apps, upload it under Settings → Capabilities → Skills.
Why doesn't the agent use the chief-of-staff skill automatically?
The SKILL.md sets disable-model-invocation to true, so the agent will not load it on its own. You need to invoke it deliberately and state your goal.
Is the chief-of-staff skill safe to use?
The SKILL.md contains no executable scripts, but its instructions push the agent to make broad environment changes. Supervise the session and review any changes to lint rules, logs or project conventions.
Who should pick a different tool instead?
Choose something else for quick one-off tasks, or for projects where the agent should not write files or spawn background work without close supervision.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use chief-of-staff if you are running a multi-day or multi-session engineering goal and want the agent to keep a written, dated record of decisions and lessons, or if you are experimenting with multi-agent workflows. Pick a simpler approach for small jobs, or for projects where you do not want the agent writing files, spawning background work or changing the environment without close supervision.
Download the skill