The Claude Code skill

A skill is a folder Claude Code reads on demand: SKILL.md plus a few scripts. The coauthr skill teaches your agent to render, publish, drain feedback, and reply — using a token that acts as you.

1. Sign in from your terminal

curl -fsSL https://coauthr.co/login | sh

Opens coauthr.co/connect to authorize this terminal (or type the code it shows) — no account needed first, the page creates one. The skill lands in ~/.claude/skills/coauthr/ and ~/.agents/skills/coauthr/ (Codex, Gemini, Copilot); the token in ~/.config/coauthr/token. Nothing outside your home directory, no sudo. Have Node? pnpm dlx coauthr.co login does the same.

2. For CI, servers, or other agents: a token

Sign in, create one in Settings ▸ Claude Code ▸ For servers and automation (shown once), then:

COAUTHR_TOKEN="$COAUTHR_TOKEN" sh -c 'curl -fsSL https://coauthr.co/skill/install.sh | sh'

Prefer to look first? Download the skill · read install.sh.

3. Use it

• “Publish to coauthr.” — the agent renders, publishes, and hands you the coauthr.co link. “Make an artifact” or “write it up” for a plan, report, or design round lands here too — the skill tells Claude Code that anything written for you to read goes to coauthr, not its built-in Artifact tool.
• “Any comments on the plan?” — it drains your review and summarizes it.
• “Fold the accepted edits and republish.” — it edits the source, resolves what it folded, republishes under the same slug.

Want that guaranteed, not just suggested? Re-run the installer with --enforce: it adds a Claude Code hook that turns any built-in Artifact publish into a coauthr publish (escape hatch: touch ~/.config/coauthr/allow-artifact, or say “claude.ai artifact”).

The API, if you'd rather script it

Bearer coa_… against https://coauthr.co/api/coauthr: GET /me, POST /publish, GET /comments?project=&status=, POST|PATCH /comments, POST|PATCH /edits. The scripts in the skill are thin curl wrappers around exactly these.