Doc Coauthoring

What it does

Turns Claude into a structured writing partner for long-form documents. When activated, Claude offers a three-stage workflow: it interviews the user to gather context, brainstorms 5-20 options per section and drafts them one at a time using surgical edits, then tests the finished doc against a fresh Claude instance (or asks the user to) to catch assumptions the authors didn’t notice. Most useful when the document has real stakes — a proposal, RFC, or spec that other people will actually read and act on — rather than a quick note.

Use cases

• Draft a decision doc or RFC where the core proposal has trade-offs and stakeholder concerns that need to be surfaced before writing prose.
• Write a technical spec or PRD section by section, with brainstorming and curation for each part instead of one long generated draft.
• Stress-test an existing document by feeding it to a context-free Claude and seeing which questions readers would ask that the doc fails to answer.

When to use it

• You’re starting a document with real audience impact (proposal, spec, decision doc, RFC).
• You have a lot of scattered context — Slack threads, meeting notes, prior docs — that needs to be consolidated.
• You want the finished doc to hold up when someone unfamiliar with the project reads it cold.
• You’re editing a shared doc and want images checked for missing alt-text before others try to parse it with Claude.

When NOT to use it

• You need a quick paragraph, email, or comment — the three-stage workflow is overhead you don’t need.
• You already have a clear outline and just want prose generated; the brainstorm-and-curate loop will feel slow.
• The document is purely personal notes with no external reader — reader testing adds no value.
• You’re writing code documentation inline (docstrings, README snippets); use a code-focused skill instead.

Installation

/plugin install doc-coauthoring@anthropic-agent-skills

If you haven’t added the Anthropic marketplace yet, run this first:

/plugin marketplace add anthropics/skills

After install, just mention it in conversation:

Help me write a decision doc about migrating our billing service from Stripe to Adyen.

Troubleshooting

Claude keeps asking clarifying questions and won’t start drafting. Tell it explicitly that context gathering is done and you want to move to Stage 2. The skill waits for that signal — if you keep info-dumping, it will keep asking follow-ups.

Reader testing isn’t running automatically. Sub-agent testing only works in environments like Claude Code where Claude can spawn a fresh instance. In claude.ai you’ll get manual instructions instead — open a new conversation, paste the doc, and ask the predicted questions yourself.

Claude rewrites the whole document on every edit instead of making surgical changes. Ask it to use str_replace and only touch the specific text you called out. The skill instructs this behavior but occasionally regresses on long sections; a one-line reminder gets it back on track.

The workflow feels too slow for the doc you’re writing. Say so — the skill is designed to yield agency. Ask to skip straight to drafting, or to work freeform. It will drop the staged process rather than push back.