Skip to content
SkillModel-led3,207 tokens in use

doc-coauthoring

A three-stage workflow for writing docs with Claude: gather context, build each section through brainstorming and edits, then test it on a fresh Claude reader.

197,763 stars on the repositoryanthropics/skills · skills/doc-coauthoring
/plugin install example-skills@anthropic-agent-skills
View source

Install command as written in the repository's README

Skill label

measured from its files

What installing doc-coauthoring costs

77tokens idle
3,207tokens in use

Idle is what sits in context while the skill is installed. In use is its instructions, loaded when it runs. Counts are approximate, made offline with cl100k_base.

Model dependence Model-led

Script-ledMixedModel-led

Mostly instructions, so how well it works depends on the model that follows them.

This describes the skill's structure, how much of it is scripts. It does not say how well it works.

Touches

Nothing detected: no shell, network, file writes, git or credentials in its files.

Declared tags come from the skill's frontmatter; detected ones from its scripts and code blocks. This is detected, not audited.

How it scores

Quality

How well the skill is written, counted from its own files: its frontmatter, its steps, examples, stated output and constraints, its size and whether every file it ships is one it names.How we score it →
Excellent
  • Frontmatter12/15
  • Structure25/25
  • Examples3/15
  • Output15/15
  • Size15/15
  • Resources10/15

Security

Starts at 100 and loses points for each thing in the skill's files that a reader should look at first.How we score it →
Excellent

No findings.

Repository health

Pre-computed score based on development activity, maintenance, community, maturity, and trend momentum.How we score it →
Good
  • Development activity72
  • Maintenance32
  • Community68
  • Maturity36
  • Momentum40

doc-coauthoring is Anthropic’s agent skill for writing documentation, proposals, technical specs, decision docs, PRDs, RFCs and similar structured content together with Claude. When a user starts a substantial writing task, Claude offers a structured workflow and, if the user accepts, acts as an active guide through three stages: Context Gathering, Refinement and Structure, and Reader Testing. A user who declines works freeform.

The skill closes the gap between what the author knows and what Claude knows before anything is drafted, then builds the document one section at a time: clarifying questions, 5-20 brainstormed options, the user’s keep, remove or combine choices, a gap check, a draft and surgical edits. Before the document goes to other people, it is tested on a fresh Claude with no context from the conversation, to catch the blind spots that make sense to the authors but confuse readers.

What You Get

  • A trigger that offers the workflow when a user mentions writing a doc, proposal, spec, PRD, design doc, decision doc or RFC, with a freeform fallback if they decline
  • A context-gathering stage with five meta-questions, an encouraged info dump and 5-10 numbered clarifying questions that can be answered in shorthand
  • A per-section loop: clarifying questions, 5-20 brainstormed options, keep/remove/combine curation, a gap check, drafting and surgical refinement
  • A document scaffold with placeholder text for every section, created as an artifact or as a markdown file in the working directory
  • A reader-testing stage that runs predicted reader questions against a fresh Claude, via sub-agents where available or by hand in claude.ai
  • A final whole-document review for flow, redundancy, contradictions and generic filler, plus closing tips on appendices and updating from reader feedback

Common Use Cases

  • Drafting a technical spec or design doc from scattered notes, chat threads and shared documents
  • Writing a decision doc or proposal where the core recommendation has the most unknowns and needs to be worked out first
  • Following a team’s existing template for a PRD or RFC while filling each section with the right context
  • Checking that a finished document answers readers’ questions before it is circulated, including readers who paste it into Claude

What Makes It Unique

Where frontend-design and pdf in the same repository shape Claude’s output for interfaces and files, doc-coauthoring shapes the conversation itself: it is a facilitation script that tells Claude when to ask, when to brainstorm and when to wait for the user, and it treats the user’s numbered selections and direct feedback as a way to learn their style for later sections. Its distinctive step is Reader Testing, where the document is handed to a Claude with no context from the session, so the doc is judged by someone who only has the page, much as a colleague would be.

Under The Hood

The skill is a single SKILL.md of instructions with no scripts or reference files. When it loads, Claude follows the staged workflow in conversation. For drafting it creates a scaffold with placeholder text, either as an artifact with create_file or, without artifacts, as a markdown file in the working directory such as decision-doc.md, and edits each section in place with str_replace rather than reprinting the document. When the user points to team channels or shared documents, it uses whatever integrations are connected, such as Slack, Teams, Google Drive or SharePoint, and asks before searching connected tools for unknown projects; without them it asks the user to paste content or enable connectors. For reader testing in Claude Code it invokes sub-agents with only the document and each question; in claude.ai it gives the user steps to test in a fresh conversation. It runs no commands of its own.

Join founders building with open source

Opinionated takes, migration guides, cost-saving tips, and insights from the open source ecosystem.

Subscribe on SubstackJoin 750+ subscribers