Home / All Skills / Content & Media Creation / doc-coauthoring
English (EN)简体中文 (ZH)日本語 (JA)
Content & Media Creation · anthropics/skills

doc-coauthoring

Guide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks.

L2 Local · Restricted workspaceV1 Commit-PinnedWrite FileDual-pass Audit
Author anthropicsGitHub anthropics/skills ↗Stars 174801License 未声明 —— doc-coauthoring/ 目录内无 LICENSE.txt（同仓 18 个 skill 中多数带 Apache-2.0 或 Anthropic 自定义 license），仓库根也无 LICENSE（gh api license 字段为 null）Commit 41bbe19d1a
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

The claimed 'guided co-authoring' relies on no code whatsoever, but purely on prompt orchestration: the model drives the user as an active guide following a three-phase interaction protocol (Context Gathering / Refinement & Structure / Reader Testing), with capabilities implemented entirely via structured conversational instructions.

doc-coauthoring/SKILL.md:L8
This skill provides a structured workflow for guiding users through collaborative document creation. Act as an active guide, walking users through three stages: Context Gathering, Refinement & Structure, and Reader Testing.
The entire skill consists of just one file, SKILL.md (`git ls-files doc-coauthoring/` returns only this file), with no scripts, imports, or templates. Here is 'why it works': it encodes document co-authoring into an executable conversational protocol rather than relying on external toolchains.

Stage 1 first transfers the 'knowledge in the user's head' into conversational context: 5 meta-questions + encouraging freeform brain dumps + generating 5-10 clarification questions based on gaps, using determinable exit criteria to prevent writing prematurely.

doc-coauthoring/SKILL.md:L30
**Goal:** Close the gap between what the user knows and what Claude knows, enabling smart guidance later.
Supporting evidence: L34-40 five meta-questions; L92-94 'Generate 5-10 numbered questions based on gaps in the context.'; L97 'Sufficient context has been gathered when questions show understanding - when edge cases and trade-offs can be asked about without needing basics explained.'—the exit criteria operationalize 'adequate understanding,' serving as the prerequisite for asking sharp, targeted 5-10 questions in each subsequent section.

Stage 2 breaks drafting down into a fixed 6-step cycle per section (clarification -> brainstorm 5-20 items -> numbered curation -> gap check -> draft -> iterate), turning vague user feedback into executable instructions via a 'numbered selection' protocol.

doc-coauthoring/SKILL.md:L168
Generate 5-20 numbered options based on section complexity. At the end, offer to brainstorm more if they want additional options.
Supporting evidence: L175-179 curation examples 'Keep 1,4,7,9', 'Combine 11 and 12'; L183-184 gap check. The numbering protocol enables the model to reliably ingest user feedback and naturally learn prioritization regarding 'what should be kept.'

Drafting and iteration entirely utilize host editing tools (`create_file` to establish the artifact scaffold / `str_replace` for surgical edits), deliberately maintaining 'never reprint the whole doc' to prevent drift caused by repeatedly rewriting long documents in full.

doc-coauthoring/SKILL.md:L208
- Use `str_replace` to make edits (never reprint the whole doc)
Supporting evidence: L135 'Use `create_file` to create an artifact. This gives both Claude and the user a scaffold to work from.'; L148 (when in a non-artifact host) 'Create file with all section headers and placeholder text.'. The skill automatically routes based on host capabilities: uses `create_file` + provides links after each edit if artifacts exist; otherwise, saves to a local markdown file.

Stage 3 Reader Testing is the skill's most distinctive mechanism: using a 'zero-context fresh Claude sub-agent' to simulate real readers, verifying whether the document remains self-contained outside the author's head—turning the 'reader's perspective' into an automated check.

doc-coauthoring/SKILL.md:L244
**Goal:** Test the document with a fresh Claude (no context bleed) to verify it works for readers.
Supporting evidence: L265 'For each question, invoke a sub-agent with just the document content and the question.'; L273 the sub-agent further checks ambiguity/false assumptions/contradictions; upon failure, 'Loop back to refinement for any problematic sections.' (L325). Environments lacking sub-agents (claude.ai web) degrade to providing manual testing instructions for the user: open a new session at https://claude.ai and paste the document (L303-304).

User agency is preserved throughout: can be declined at the start, skipped mid-way, or pivoted to freeform anytime; the process guides without railroading, maintaining clear boundaries that 'the user owns the document.'

doc-coauthoring/SKILL.md:L26
If user declines, work freeform. If user accepts, proceed to Stage 1.
Supporting evidence: L358 'If user wants to skip a stage: Ask if they want to skip this and write freeform'; L360 'Always give user agency to adjust the process'; L338 wrap-up requires the user to do the final read and take responsibility ('they own this document and are responsible for its quality').

2Core Capabilities

01Doc task trigger detection + proactive offer of a 3-phase workflow (including a freeform opt-out path)
02Stage 1 context gathering: 5 meta-questions -> freeform info dump -> 5-10 gap-driven clarification questions (supports shorthand answers)
03Document scaffolding generation: uses `create_file` if artifacts exist, or creates placeholder markdown in workspace (decision-doc.md / technical-spec.md, etc.)
04Section-by-section 6-step co-authoring protocol: clarification questions -> 5-20 item brainstorming -> numbered curation -> gap check -> `str_replace` drafting -> iteration
05Style learning feedback loop: prompts user to 'point out what to change' rather than editing directly, allowing the model to learn preferences for subsequent sections
06Convergence quality gates: checks if content can be trimmed after 3 rounds without substantive changes; full review for flow/redundancy/contradictions/slop at 80%+ completion
07Dual-path Reader Testing: automated testing via sub-agents in Claude Code / manual test script for claude.ai web; problematic sections loop back to refinement
08Wrap-up post-processing: author takes final responsibility, co-authoring chat link added to appendix, doc continuously updated with real reader feedback

3External Dependencies

KindDependency / Endpoint
apicreate_file（宿主内建 artifact 工具，非外部 API）
apistr_replace（宿主内建编辑工具，非外部 API）
apisub-agents（Claude Code 运行时能力，非外部 API；skill 未指定具体 task 工具名）
api集成连接器 Slack/Teams/Google Drive/SharePoint/MCP 服务器 —— 条件性：仅在『If integrations are available』时提及，由宿主提供、无绑定端点
networkclaude.ai —— 仅作为用户手动 Reader Testing 的文字入口（用户自行打开新会话，skill/agent 不发起网络调用）

4Recommended Governance & Evidence

Suggested Controls for Safe Operation
For Agent Runtime
Restrict working directory
For Administrators
Retain operation logs
Evidence & Triggers (File Paths & Excerpts)
  • 本地工作目录: File read/write and artifact output: performs `str_replace` edits on co-authored documents — SKILL.md L188/L208
Audit Notes: Pure instructions with no scripts, network exfiltration, or credential access; however, the skill explicitly directs the host to perform local file writes (L148 scaffold file creation, L188/L208 `str_replace` edits on the user's document). Per the risk paradigm 'local read/write only, no network exfiltration, no credentials,' it falls under 🔵 Blue Tier ('Notice / Informational'). The write targets are always the co-authored documents progressively confirmed through user conversation, with no covert behavior.

5Second-Pass Verification

  • [ok] 目录内仅 SKILL.md、无脚本/引用/模板等资源 — git ls-files doc-coauthoring/ 只返回 doc-coauthoring/SKILL.md
  • [ok] description 引用逐字一致 — L3 原文与 meta.official_desc 完全一致
  • [ok] create_file / str_replace 调用点真实存在 — L135、L188、L208、L367-368 均实存，且 L367-368 收尾 Tips 再次强调同一工具对
  • [ok] claude.ai 网络提及 — 全文件唯一 http(s) 命中为 L303；L72/L290 仅文字提及 Claude.ai 产品名，无调用语义
  • [ok] Slack/Teams/Google Drive/SharePoint/MCP 集成语 — L70 实存且为条件式（If integrations are available），属宿主能力提示而非 skill 自带 API
  • [ok] sub-agent Reader Testing 机制 — L251-253（无需用户参与的自动路径）、L265、L273 实存，L290 起手动路径对照完整
  • [ok] 无凭证读取 — grep api key/token/secret/password/credential/env/export/keychain 全文件零命中
  • [ok] 无脚本执行面 — grep bash/python/subprocess/curl/wget/exec( 全文件零命中；SKILL.md 内无代码块

6Verdict & Scenarios

  • Process as product: breaks doc co-authoring into section-by-section, determinable interaction protocols (6 steps per section + exit criteria), rather than a one-shot 'write me a doc'
  • Reader Testing uses zero-context Claude as a real reader to test-read, directly verifying document usability outside the author's context—a validation loop absent in most writing skills
  • Style learning loop: guides users to 'point out what to change' rather than editing directly, letting the model learn preferences on the fly and align better in later sections
  • Environment-adaptive dual paths: executes all three phases seamlessly in both Claude Code (automated sub-agent testing) and claude.ai web (manual guidance)
  • Restrained without overstepping: model acts purely as a guide; document ownership, final read, and fact-checking remain firmly with the user, complete with built-in anti-slop checks
    • Best for: Ideal for users who already have raw materials and want to turn PRDs, design docs, decision records, or RFCs into structured documents that 'remain self-contained outside the author's head and can be consumed by teammates or Claude'; highest value in Claude Code (sub-agents and artifacts available), where Reader Testing is fully automated. Writers willing to engage in multi-turn Q&A and curation will benefit the most.
    Not for: Users seeking one-shot instant drafts or simple templates/examples; those lacking patience for the Q&A-curate-iterate loop; scenarios where doc content is highly confidential and cannot leave the current conversation (e.g., handed to a sub-agent or pasted into claude.ai).
    Install Agent-Ready
    ① Verified Mirror 2026-09-06
    Option A · Direct DownloadDownload doc-coauthoring.tar.gz
    sha256: 930944490c8451b9…
    Option B · Copy Guide for Agent
    install.json
    Agents automatically download and verify mirrors from this JSON guide.
    ② Upstream GitHub
    GitHub Repo ↗
    Pinned to commit 41bbe19d1a.