Kipt

The doc-coauthoring skill: write specs and proposals with Claude

Anthropic's doc coauthoring skill walks you and Claude through context gathering, section-by-section drafting and reader testing for specs, PRDs and proposals.

By the Kipt team · Updated 24 October 2026

Source: anthropics/skills · Maintainer: Anthropic · License: Not stated (no LICENSE file in the folder or the repo) · Checked: 24 October 2026

What is the doc-coauthoring skill?

The doc-coauthoring skill turns Claude into a guide for writing a structured document with you: it gathers context first, builds the document one section at a time with your feedback, then tests the draft on a fresh Claude that has never seen your conversation. It is for engineers and product people writing specs, PRDs, RFCs, design docs, decision docs and proposals that other people (and their AI assistants) will read.

Its description in the SKILL.md begins: "Guide users through a structured workflow for co-authoring documentation."

It lives in Anthropic's anthropics/skills repo. The skill's skills.sh page listed 93.9K installs (snapshot, 24 October 2026).

What does the doc-coauthoring skill do?

The SKILL.md lists its triggers: phrases like "write a doc", "draft a proposal", "create a spec" or "write up", doc types like "PRD", "design doc", "decision doc" or "RFC", or a user who "seems to be starting a substantial writing task." Claude then offers the workflow and asks whether you want it or prefer freeform.

Stage 1: Context gathering

The goal is to "close the gap between what the user knows and what Claude knows." Claude asks five meta questions: the document type, the primary audience, the impact you want, any template, and other constraints. Then it invites an unorganized dump of background, team discussions, rejected alternatives, timelines, architecture and stakeholder concerns. If connectors such as Slack, Google Drive or SharePoint are available, it can pull context from them. After the dump it asks 5 to 10 numbered clarifying questions, which you can answer in shorthand like "1: yes, 2: see #channel."

The stage ends when Claude can ask about edge cases and trade-offs "without needing basics explained."

Stage 2: Refinement and structure

Claude agrees the sections with you (suggesting 3 to 5 if you don't have a structure), creates a scaffold with placeholders, and starts with the section that has the most unknowns (the core proposal for a decision doc, the technical approach for a spec). For each section it:

  1. Asks 5 to 10 clarifying questions.
  2. Brainstorms 5 to 20 numbered options.
  3. Asks which to keep, remove or combine ("Keep 1,4,7,9").
  4. Checks for gaps.
  5. Drafts the section.
  6. Refines it with targeted edits, never reprinting the whole doc.

It asks you to say what to change rather than editing directly, so it learns your style for later sections. Near the end it rereads the whole document for flow, contradictions and "anything that feels like 'slop' or generic filler."

Stage 3: Reader testing

Claude predicts 5 to 10 questions a reader would ask, then checks whether a fresh Claude, given only the document, answers them correctly. It also asks that reader to flag ambiguity, assumed knowledge and contradictions. Gaps send you back to Stage 2. The doc is ready when the reader "consistently answers questions correctly."

What is in the folder?

One file: SKILL.md, about 16 KB. No scripts, templates or license file.

How do you install the doc-coauthoring skill?

Three documented routes. Pick one per agent.

Claude Code plugin. The anthropics/skills README registers the repo as a plugin marketplace. In a Claude Code session:

/plugin marketplace add anthropics/skills
/plugin install example-skills@anthropic-agent-skills

That installs the whole example-skills set (12 skills in the repo's marketplace file). Plugin skills are namespaced, so this one appears as /example-skills:doc-coauthoring. See Claude Code plugins.

The skills CLI. To install only this skill, use the command from its skills.sh page:

npx skills add https://github.com/anthropics/skills --skill doc-coauthoring

The skills CLI README says it installs into the current project by default; add -g for all projects or -a claude-code to target Claude Code. The folder lands in .claude/skills/doc-coauthoring/ or ~/.claude/skills/doc-coauthoring/, the Claude Code skill locations. Codex and Cursor read the same SKILL.md format; see Codex skills and Cursor skills.

Claude.ai. The README says the example skills "are all already available to paid plans in Claude.ai," so there is nothing to upload. The help article on using skills in Claude covers turning skills on under Customize, Skills, and notes that skills need code execution enabled.

How do you use it?

Start a writing task and accept the offer. Some prompts that fit (suggestions, not quotes from the repo):

  • "Help me write a design doc for moving our job queue from cron to a managed queue."
  • "I need a decision doc: Postgres or DynamoDB for the new events service. Audience is the platform team."
  • "Draft an RFC for deprecating API v1. I'll dump the context first."
  • "Write a PRD for team workspaces. Use the template in docs/prd-template.md."
  • "Here's my draft spec. Run reader testing on it and tell me what a new engineer would get wrong."

Where does it work best?

The skill adapts to where it runs:

  • Claude Code: with no artifacts, it creates a Markdown file in your working directory (for example decision-doc.md) and edits it in place. Reader testing runs automatically with subagents, each getting only the document and one question.
  • Claude.ai: it drafts in an artifact, can pull context through connectors you've enabled, and walks you through reader testing manually in a fresh conversation.

No packages or API keys are needed. Connectors are optional and only matter if you want Claude to read your team's channels or documents.

Tips and limits

  • It is slow on purpose. Three stages with numbered questions take longer than "write me a spec." The skill tells Claude to suggest ways to move faster if you seem frustrated, and you can skip any stage.
  • Answer in shorthand. It expects terse replies: numbers, "keep 2, 5, drop the rest." That keeps the loop quick.
  • Reader testing catches assumed context. The fresh reader sees only the doc, so anything you explained in chat but never wrote down shows up as a wrong answer.
  • You still own the doc. The final review tells Claude to recommend a read-through and a check of "any facts, links, or technical details."
  • Third-party content. On 14 September 2026, Snyk's audit on skills.sh rated it Warn (medium risk) for third-party content exposure, because the workflow can read shared channels and documents through connectors, an indirect prompt injection risk. Gen Agent Trust Hub and Socket showed Pass as of 24 October 2026.
  • No license stated. The other 18 skill folders in the repo each ship a LICENSE.txt; this one doesn't, and the README only says "many" skills are Apache 2.0. GitHub's licensing docs explain that without a license, default copyright law applies and no one may reproduce, distribute or create derivative works. Install it from Anthropic's repo and don't republish or share copies. Anthropic's README also calls the example skills "provided for demonstration and educational purposes only."

Save it in your Kipt library

Kipt can import a skill when you paste its GitHub folder link into Skills, but skip the import for this one. With no license stated, Anthropic hasn't granted the right to copy or modify it, so install it from Anthropic's repo with the plugin or skills CLI commands above. Kipt is a better home for workflow skills you write yourself in the same style: they live in your own private GitHub repo and install into Claude Code or Codex with one command:

npx skills add https://github.com/<you>/<repo>.git --all -g

Kipt is not affiliated with Anthropic.

Similar skills

Questions

What is the doc-coauthoring skill?

It is an Agent Skill from Anthropic, published in the anthropics/skills GitHub repo, that runs a structured workflow for writing documents such as technical specs, PRDs, RFCs, decision docs and proposals. It has three stages: context gathering, section-by-section refinement, and reader testing with a fresh Claude.

What is reader testing in the doc-coauthoring skill?

It is the third stage. Claude predicts 5 to 10 questions a reader might ask, then has a fresh Claude with no context answer them using only the document. In Claude Code it does this with subagents; on claude.ai it gives you steps to run the test yourself in a new conversation.

How do I install the doc-coauthoring skill?

In Claude Code, run /plugin marketplace add anthropics/skills, then /plugin install example-skills@anthropic-agent-skills. To install only this skill, run npx skills add https://github.com/anthropics/skills --skill doc-coauthoring. On paid Claude.ai plans, Anthropic's README says the example skills are already available.

Can I skip the workflow and just write the doc?

Yes. The skill tells Claude to offer the workflow and ask whether you want it or prefer to work freeform. If you decline, Claude writes freeform. You can also skip a stage later, and the skill says to always give you agency to adjust the process.

What license is the doc-coauthoring skill under?

None is stated. It is the only one of the 19 skill folders in the repo without a LICENSE.txt, its frontmatter has no license field, and the repo has no top-level license file. Without a license, default copyright applies, so install it from Anthropic's repo and do not republish or share copies unless Anthropic grants permission.

More in Skill directory

All guides

Keep your prompts and skills in one place, in your own GitHub. Install every skill with one command.

Get started

Kipt is an independent product, not affiliated with or endorsed by Anthropic, OpenAI, Cursor or GitHub. Claude, ChatGPT, Codex, Cursor and GitHub are trademarks of their owners.