technical-writer for Claude Code: a skill with writing steps for each document type

technical-writer is a skill that gives an agent writing steps and quality checks for each type of document. This post covers the problem it solves, who it suits, how to set it up and trigger it, and the one thing to change before you adopt it.

What problem it solves

Ask an agent to “write this up” and the shape of the result depends on the day. A research note comes back as an essay, a README skips the install step, and a social post runs twice as long as the channel allows. The writing is often fine. What is missing is a fixed idea of what this kind of document must contain and in what order.

technical-writer fixes that idea before any drafting starts. It confirms the reader and the goal first, then picks a type-specific playbook, such as research notes, API documentation, or operating procedures. It drafts short sections from a minimal outline, each with one concrete example, and the document only counts as finished after it passes five checks: accuracy, completeness, clarity, structure, and actionability. It is one of the four content marketing skills from my own workflow, which is why it has no public repository.

Who it fits

The chapter rates four situations. Three are enough to decide.

Situation	Fit	Reason
Saving research results as notes	Good fit	Fixed fields for measured values and sources
Drafting for two social channels from the same topic	Good fit	Length and tone rules for each channel
Sharing directly through Google Docs	Poor fit	Uploading belongs to another skill

Internal API documentation and operating procedures sit in the middle. The skill covers them, but only with a brief item order.

Setup and first run

There is no repository to install from. You build your own skill with this layout.

# Example layout in the user skill folder
~/.claude/skills/
  technical-writer/
    SKILL.md

It activates when you include its name in a request to write or polish a document.

Write [type of document to create]

The part that pays off most is the link between research and distribution. A research note states its one-line summary as a claim of the form “X is Y,” lists measured data, insights, and sources, and forbids estimates and placeholder values. The same note format carries fields for a LinkedIn draft and a Threads draft, each with its own length and tone limits, so one round of research turns into two channel drafts without a second pass. Vocabulary also shifts with the reader, with separate tone and terminology levels for developers, nondevelopers in training, and social media readers.

What trips people up

The research note destination is set to the author’s environment. The instructions save notes into a specific folder of a research repository, with date-prefixed file names. If you copy the skill as is and have no such repository, the agent will try to write into a path that does not exist on your machine. Change the destination to a folder you actually keep before the first run, and while you are in the file, check the LinkedIn hashtag guidance against your own channel’s style, since the skill itself notes that it may differ.

Where this comes from

This post adapts part of chapter 02-06 of The Claude Code & Codex Skills Guidebook. The full chapter covers the full component table, including the four technical document types and their step sequences, and the clauses that reference a retired skill and how to replace them, and the book has 52 more skills like this one. Read the full chapter in The Claude Code & Codex Skills Guidebook on Gumroad.