Doc Co-Authoring

Doc Co-Authoring is the official Anthropic Agent Skill that turns Claude into an active guide for writing documentation — proposals, technical specs, PRDs, RFCs, decision docs, and similar structured content. It runs a structured three-stage workflow that transfers your context to Claude, builds the document section by section, and then tests the finished doc on a fresh Claude that has no prior context, catching the blind spots that confuse real readers.

What is Doc Co-Authoring?

It is a skill in Anthropic's official anthropics/skills repository that gives Claude a repeatable process for collaborative writing instead of one-shot drafting. Rather than dumping a generic draft, Claude walks you through three stages — Context Gathering, Refinement & Structure, and Reader Testing — so the document reflects what you actually know and works when someone else reads it (including when they paste it into Claude).

What it does

• Offers a structured co-authoring workflow and explains the three stages before starting
• Stage 1 — Context Gathering: asks meta-questions (doc type, audience, desired impact, template), invites a stream-of-consciousness info dump, then asks 5–10 clarifying questions to close knowledge gaps
• Stage 2 — Refinement & Structure: builds each section by asking clarifying questions, brainstorming 5–20 options, letting you curate keep/remove/combine, drafting, and refining with surgical str_replace edits
• Stage 3 — Reader Testing: predicts the questions readers will ask and tests them against a fresh Claude (a sub-agent in Claude Code, or a manual second conversation on Claude.ai) to surface ambiguity, false assumptions, and contradictions
• Checks shared docs for images missing alt-text so reader Claude can still understand them

How it works

Install the skill and tell Claude you want to write a doc. Claude offers the workflow and, if you accept, starts with context gathering. It creates a scaffold — an artifact via create_file where artifacts are available, otherwise a markdown file in your working directory — then fills it in section by section, starting with the section that has the most unknowns (usually the core proposal or technical approach) and leaving summaries for last. Throughout, you describe changes rather than editing directly so Claude learns your style; every edit is a surgical str_replace, never a full reprint. Reader Testing then validates the doc before anyone else sees it.

Who it's for

Anyone who writes structured documents with Claude — product managers drafting PRDs, engineers writing design docs and RFCs, founders and leads writing proposals and decision docs, and teams who need a doc to land correctly with stakeholders. It works in Claude.ai, the Claude app, and Claude Code, adapting its testing approach to whether sub-agents and artifacts are available.

What you can build

• Product requirement docs (PRDs) and feature proposals
• Technical specs, design docs, and RFCs
• Decision docs and architecture decision records
• Project proposals and structured write-ups that must persuade a specific audience

Why it matters

Most AI writing fails because it skips the two hardest parts: getting your real context into the model and verifying the result lands for someone without that context. This skill makes both explicit — a disciplined intake up front and an adversarial reader test at the end — so the document is not just well-written but actually works when others read it. The result is fewer review cycles and docs that hold up on their own.

1. Install with the skills CLI

npx skills add anthropics/skills --skill doc-coauthoring

2. Or add it as a Claude Code plugin

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

3. Use it

Tell Claude what you want to write, for example: "Help me write a decision doc about our database migration." Claude will offer the structured co-authoring workflow.