Technical Writing

Reviewed 2026-10-14writingPrompt / Rule# technical-writing # documentation # specs # design-docs # writing

Anthropic's doc-coauthoring skill: a three-stage workflow in which the agent interviews you, drafts a spec or design doc section by section, then tests it on a fresh reader.

Security Reviewed

Very low risk: a single Markdown workflow with no scripts, commands or network calls of its own.

Install Technical Writing in Google Antigravity

Antigravity loads skills from .agents/skills/ in your project or ~/.gemini/config/skills/ for all projects (official docs). Full guide

EasiestOption 1: paste this to your Antigravity agent

Install the "Technical Writing" skill for Google Antigravity.
Follow the steps at https://www.antigravityskills.org/install/technical-writing.md (source: https://github.com/anthropics/skills/tree/main/skills/doc-coauthoring).
Before running anything, show me what you will download and whether it includes scripts, and wait for my OK.
Install it into this project (.agents/skills/) unless I say global. When done, check that Antigravity lists it as an installed skill.

The agent reads the install file, shows you what it will download and waits for your OK.

Option 2: install it yourself

1. Download the repository
```
git clone --depth 1 https://github.com/anthropics/skills /tmp/anthropic-skills && git -C /tmp/anthropic-skills log -1 --format=%h -- skills/doc-coauthoring
```
The second command prints the last commit that changed this skill. We reviewed the folder at 683bc88; a different hash means the skill changed after our review.
2. Copy the skill into this project (workspace skill)
```
mkdir -p .agents/skills && cp -r /tmp/anthropic-skills/skills/doc-coauthoring .agents/skills/
```
Run from your project root. The folder contains one file, SKILL.md.
3. Or install it for all projects (global skill)
```
mkdir -p ~/.gemini/config/skills && cp -r /tmp/anthropic-skills/skills/doc-coauthoring ~/.gemini/config/skills/
```
Use this instead of the previous step. For the Antigravity CLI the global folder is ~/.gemini/antigravity-cli/skills/.
4. Check that Antigravity sees it
Antigravity 2.0 app: start a new conversation and ask "What skills do you have?"; doc-coauthoring should be listed. Antigravity CLI: run /skills (/skills reload in a session that was already open).

The anthropics/skills README documents only the Claude Code plugin marketplace (and Claude.ai / the API) and does not mention Antigravity. These steps copy the plain SKILL.md folder into Antigravity's standard skill folders. No API keys or binaries needed.

Using Claude Code, Cursor or Codex? See other assistants below.

What is Technical Writing?

doc-coauthoring is Anthropic's skill for engineers, product managers and anyone who has to write a technical spec, design doc, decision doc, PRD, RFC or proposal with an agent. Instead of writing a draft in one go, the agent works with you in three stages: Context Gathering (it asks about doc type, audience and desired impact, lets you dump context, then asks 5-10 numbered clarifying questions), Refinement & Structure (for each section: questions, 5-20 brainstormed options, your keep/remove/combine choices, a gap check, a draft and edits), and Reader Testing (a reader with no context answers predicted questions from the doc so gaps show up). What we counted at commit 683bc88 (2026-10-05; the folder is unchanged at the repository's latest commit dbd4588 on 2026-10-14): one file, SKILL.md, 375 lines / 2,466 words / 15,815 bytes, with no scripts and no frontmatter beyond name and description. Stage 2 has 6 steps; Stage 3 has two paths, one with sub-agents and one where you test manually in a new chat.

How to use Technical Writing?

Copy the folder into Antigravity's skills folder (steps below), then say you want to write a doc, for example "draft a design doc" or "write up an RFC". The skill first offers the structured workflow and works freeform if you decline. Without an artifact tool it creates a Markdown file in your working directory (for example technical-spec.md) with every section header and placeholder text, then fills it in section by section with targeted edits rather than reprinting the whole document. When about 80% of the sections are done it re-reads the whole document for flow, redundancy and filler.

Key features of Technical Writing

✓5 meta questions up front: doc type, audience, desired impact, template, constraints
✓5-10 numbered clarifying questions you can answer in shorthand ("1: yes, 2: see #channel")
✓Per section: clarifying questions, 5-20 brainstormed options, curation, gap check, draft, iterative edits
✓After 3 iterations with no substantial change it asks what can be cut
✓Reader Testing with a context-free sub-agent, or a manual test plan for a fresh chat
✓Reads Slack, Teams, Google Drive or SharePoint only through integrations you have connected, and asks before searching for unknown names

When to use Technical Writing

1Write a technical spec or design doc for a new feature
2Draft an RFC or decision doc for team review
3Turn scattered notes and thread context into a structured proposal
4Check whether a finished doc answers the questions readers will ask

Installation & Setup Guide

Agent Rules / Prompt Setup

Follow these steps (taken from the project's own docs) to configure Technical Writing in your development environment.

Step 1: Claude Code: add the marketplace (README)claude

/plugin marketplace add anthropics/skills

Run inside Claude Code.

Step 2: Claude Code: install the example-skills plugin (README)claude

/plugin install example-skills@anthropic-agent-skills

doc-coauthoring is one of the 12 skills in the example-skills plugin (.claude-plugin/marketplace.json), so this installs all 12.

Step 3: Cursor: not documented by the projectcursor

The README covers Claude Code, Claude.ai and the API only, so we do not list a Cursor path.

Usage Example & Verification

Prompt / Request Example

1) README-style trigger (from the skill's trigger list): "Draft a design doc for our new rate limiter."
-> What happens: the agent offers the three-stage workflow, then asks the 5 meta questions (type, audience, impact, template, constraints) before writing anything.

2) Our example: "Help me write a technical spec for moving our job queue to a new service. The audience is the platform team. Here are my notes: ..."
-> What happens: after your context dump the agent asks 5-10 numbered questions, creates technical-spec.md with section headers and placeholders, and then works one section at a time: brainstorm a numbered list, you say what to keep, it drafts.

3) Our example: "The spec is done. Run reader testing on it."
-> What happens: the agent predicts the questions a reader would ask and has a fresh sub-agent answer them from the document alone. If your agent cannot start a sub-agent, the skill gives you the questions to try in a new chat instead.

Installation Confirmation Method

Ask "What skills do you have?" in a new Antigravity conversation (CLI: /skills) and look for doc-coauthoring. Then say "I want to write a design doc"; the agent should offer the three-stage workflow and ask about doc type, audience and desired impact before drafting.

Security Review

Static review

Reviewed Date

2026-10-14

Audited Ref / Commit

683bc88e56f3e09ba94f7055977f3d3aa499f202

Reviewer

antigravityskills.org (static code review)

Very low risk: a single Markdown workflow with no scripts, commands or network calls of its own.

Full review: scope, findings, risks

Audit Scope:Static review of anthropics/skills at commit 683bc88 (2026-10-05; skills/doc-coauthoring unchanged at HEAD dbd4588 on 2026-10-14): skills/doc-coauthoring/SKILL.md in full (375 lines, the only file in the folder), README.md, .claude-plugin/marketplace.json.

Key Findings:Instructions only: one Markdown file, no frontmatter tool restrictions. The agent creates an artifact or a Markdown file in the working directory and edits it with str_replace. It may read shared documents or channels through connected integrations and asks before searching connected tools for unknown entities. It passes the document to sub-agents for Reader Testing. One URL (https://claude.ai) in the manual testing steps. No commands, scripts, API keys or environment variables.

Identified Operational Risks & Permissions:Context you paste or pull in from team tools ends up in the drafted document; review it before sharing. No license file in the folder.

skills/doc-coauthoring contains only SKILL.md. It tells the agent to create a document scaffold (an artifact, or a Markdown file such as technical-spec.md in the working directory) and to edit it with str_replace. If integrations such as Slack, Teams, Google Drive or SharePoint are connected, it may read the channels or documents you point to, and it asks for confirmation before searching for unknown entities. Reader Testing passes the document content to sub-agents. There are no shell commands, scripts, environment variables or API keys. The only URL is https://claude.ai in the manual testing instructions.

Operational Notes & Known Limitations

Known Limitations

•It is a co-writing workflow, not a proofreader, grammar checker or API reference generator
•We counted: the word Claude appears 27 times, and the skill uses Claude tool names (create_file, str_replace, artifacts) and links https://claude.ai for manual testing. In Antigravity the agent follows the Markdown-file path; the sub-agent reader test only works if your agent can start a sub-agent, otherwise you test in a new chat
•Pulling context from team tools needs matching integrations (MCP servers); otherwise you paste content
•No license file in the skill folder and no license field in its frontmatter; the repository README says many skills in the repo are Apache 2.0
•The README does not mention Antigravity (we searched). Difference from the grant-writing page on our site: that skill is agency-specific (NSF, NIH and others); this one is a general workflow for any structured document

Frequently Asked Questions

Is the technical writing skill safe?

In our static review of commit 683bc88 it is one Markdown file with no scripts, commands or network calls of its own. It writes one document file in your working directory and reads team tools only through integrations you connected.

How do I install it in Antigravity?

Clone anthropics/skills, copy skills/doc-coauthoring into .agents/skills/ or ~/.gemini/config/skills/, then ask "What skills do you have?" (CLI: /skills). The README does not mention Antigravity.

What kinds of documents does it handle?

The skill names technical specs, decision docs, proposals, PRDs, design docs and RFCs.

What is Reader Testing?

A reader with no context (a sub-agent, or you in a fresh chat) answers predicted questions from the doc; the agent then fixes what that reader got wrong.