Guide · Agent skill
How to use the write-manual agent skill, step by step
Updated · written with AI from the project's own documentation and checked against our catalogue data
The write-manual skill is an agent skill for Claude that follows a fixed documentation method when you ask it to write a long reference manual about one technical subject at one exact version, such as a protocol or file format. This guide explains what the skill makes the agent do, how to install it from the ZIP or the Claude apps, and how to trigger it with a plain request. After reading it, you will know what the skill needs from you before you start and what its audit expects at the end.
Quick start
- 1Download the ZIP. Download the write-manual ZIP from this page.
- 2Install the skill. Unzip it into ~/.claude/skills/ for personal use, or into a project's .claude/skills/ folder. In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.
- 3Prepare the subject. Pin the subject's version and gather its specification, schema or source tree at a pinned commit.
- 4Ask in plain language. Request a manual for the named subject and version, using a phrase such as write a 101 manual for the subject.
- 5Run the audit. Have the agent fix every finding until the audit reports zero findings.
What the write-manual skill makes the agent do
Without the skill, asking an agent for a manual usually gets one long writing pass. The write-manual skill replaces that with a documentation workflow. The agent pins the subject to one version, ranks its sources by authority, and records the real requests and responses of the subject from an offline capture. It then writes each section in a fixed shape.
The skill ends with an audit that must report zero findings. The source describes the audit as the gate that decides whether the manual is finished.
- Pins the version of the subject, such as a protocol or specification.
- Ranks sources in authority order, from schema or specification down to a capture kit.
- Records requests and responses from an offline capture rather than writing them from memory.
- Writes each section in a fixed shape with a thesis, takeaways and a sources line.
- Runs an audit that must report zero findings before the manual is accepted.
Install write-manual from the ZIP or the Claude apps
Download the ZIP for write-manual from this page. To use it with Claude Code, unzip it into your user skills folder at ~/.claude/skills/. To keep it for one project only, unzip it into the .claude/skills/ folder inside that project instead.
If you use the Claude apps, you can upload the skill instead. Open Settings, go to Capabilities, then choose Skills and upload the ZIP there.
- Personal use across projects: unzip into ~/.claude/skills/.
- One project only: unzip into the project's .claude/skills/ folder.
- Claude apps: upload the ZIP under Settings → Capabilities → Skills.
Trigger the skill with a natural request
You do not need a special command. The skill is triggered by requests that match its description. The description lists phrases such as write a manual, write a 101 manual for a subject, manual.json, add a manual section, build a capture kit, audit the manual, and pin a spec version and document it.
Name the subject and the version in your request. The skill is built around one subject at one pinned version, so a request without a version will be harder to follow.
- Write a 101 manual for the protocol at the version I have pinned.
- Build a capture kit for this runtime so the examples come from real output.
- Audit the manual and fix every finding until the audit reports zero.
- Add a manual section about the cancel operation, using the spec at my pinned commit.
What you need before you start
The skill needs material it can check. Give it access to the subject's specification, schema, or source tree at a pinned commit. It also needs a way to run the subject, or a faithful stand-in, offline so it can write capture files.
The source says Python or the repository's build tooling is needed for the audit and manual build. This is inferred from the referenced capture kit and audit files and is not confirmed by the source, so check your repository before relying on it.
- Specification, schema, or source tree at a pinned commit.
- A way to run the subject or a faithful stand-in offline.
- Python or the repository's build tooling for the audit and build (inferred, not confirmed).
- Runs scripts: yes. Uses network: no.
Safety notes: scripts and network access
The skill runs scripts. The capture kit is a program that drives the subject and writes its output to files. The source says the capture kit runs offline, needs no network and needs no keys, and that the worked example uses the Python standard library only.
Read the capture kit and the audit scripts in your repository before you run them, since the skill runs scripts as part of its method. The listing states that the skill does not use the network.
- The skill runs scripts, so review them before running.
- The listed network use is none.
- The capture kit is designed to run offline with no keys.
Tips for getting a useful manual
Keep the scope narrow. The skill is meant for a manual of about 60 to 100 pages, and it is not meant for short how-to guides. A manual of fewer than about 15 sections does not need this shape.
When the audit reports a finding, fix the text, the figure or the capture. The source says never raise a limit to pass the audit.
- Pick a subject that has a fixed version and does not change weekly.
- Use the skill for a protocol, specification or runtime, not a general concept.
- Let examples come from captures, not from memory.
- Run the audit after each section and fix findings as they appear.
node scripts/audit_manuals.js --manual <id>What people use it for
- Documenting a protocol that other engineers need to maintain.
- Building a versioned reference for a file format with examples from real captured output.
- Writing a manual for a runtime inside a repository that already has the manuals tooling.
- Adding a single section to an existing manual that follows the same shape.
Frequently asked questions
How do I install the write-manual agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use, or into the .claude/skills/ folder of a project. In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.
Is the write-manual skill free to use and what is its licence?
The listing states that the licence is MIT. The source does not state any price, so check the repository for any cost information.
What do I need to run the write-manual skill?
You need access to the subject's specification, schema or source tree at a pinned commit, and a way to run the subject or a faithful stand-in offline. Python or the repository's build tooling is likely needed for the audit and build, but this is inferred and not confirmed.
Does the write-manual skill use the internet?
No. The listing says it uses no network, and the capture kit described in the source runs offline with no keys.
What should I do if the audit does not report zero findings?
Fix the text, the figure or the capture that caused each finding and run the audit again. The method says never to raise a limit to make the audit pass.
How is write-manual different from asking an agent to write documentation?
The skill is a documentation workflow rather than a single writing prompt. It pins the version, ranks sources, records captured output and runs an audit. For short how-to guides or general concept lessons, the skill is not the right tool and a lesson or guide is better.
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 write-manual if you are an engineer or technical writer who must document one protocol, specification or runtime at a pinned version and can run that subject offline to capture its output. Pick something else for short how-to guides, general concept lessons, or subjects that change every week.
Download the skill