Imported from Capafy/Capafy-skills (AGENTS.md). Install upstream with npx skills add Capafy/Capafy-skills. Copyright stays with the author.

AGENTS.md

Guidance for AI coding agents (Claude Code, Codex, OpenClaw, Cursor, Copilot, and others) working in this repository or helping a user adopt Capafy.

What this repo is

This is the official repository for Capafy — an agent skills marketplace for Claude Code, Codex, and OpenClaw. Domain experts package their know-how into Skills; users discover, buy, and run them, either directly on https://capafy.ai or from inside their own agent client.

The repo distributes Capafy's two official Skills:

• capafy-user/ — install this to use Skills built by others: search the marketplace, buy, and run Agents from inside your normal chat.
• capafy-publisher/ — install this to publish a Skill you built and earn when others run it.

README.md is the full human-facing documentation. LICENSE is MIT.

Requirements

• One of: Claude Code, Codex, or OpenClaw
• Python 3.8+
• A Capafy account (sign up at https://capafy.ai)

How to install the Skills

In the agent client's chat, send either or both:

install https://capafy.ai/install-user-skill.md
install https://capafy.ai/install-publisher-skill.md

Verify the install by asking the client to list the capafy skills — you should see capafy-user and/or capafy-publisher. The first time either Skill is used, it updates itself and asks the user to log in.

Authentication

Login happens through the agent client: the user says log into Capafy, Capafy emails a one-time code, and the Skill saves a token locally for later runs. The same account works as both user and publisher.

Using capafy-user

Search the catalog (e.g. "find an Agent that summarizes PDFs"), order an Agent (confirm the billing plan; pay with in-client credits or by card via Stripe on the web), then describe a task and let the client route it. For Run Online Agents, a lightweight Thin Skill router is installed locally; for Download Agents, the full package installs and runs on the user's machine.

Using capafy-publisher

Point it at a Skill you've built. The flow runs through a few commands with web checkpoints in between:

publish-init        # scan workspace, list Skill candidates
publish-configure   # scan for secrets, stage the bundle (add --deep-scan for a thorough sweep)
publish-ship        # validate, package, upload
publish-status      # check where a draft stopped

Each command prints a review_url to open in the browser; a review_url expires after 1 hour (re-run the same command with the same --agent-id for a fresh one). publish-ship returning shipped only means the bundle uploaded — the user must still click Submit on the final web page for it to enter review.

Conventions for agents working in this repo

• The language is Python (100% of the repo). Target Python 3.8+.
• A publishable Skill is a directory containing a SKILL.md plus optional scripts, references, and config.
• Never commit secrets to Skill source — use environment variables. For Run Online Skills, the publisher stages credentials as platform-hosted placeholders rather than shipping them.
• --runtime-dir must be the absolute path the agent client opened the project at; it is not inferred. For OpenClaw it must be a real workspace path (e.g. /home/me/.openclaw/workspace_xxx), not ~/.openclaw.

Key links

• Website: https://capafy.ai
• Marketplace: https://capafy.ai/search
• Publisher Guide: https://capafy.ai/developer/doc
• Become a Publisher: https://capafy.ai/earn
• Help Center: https://capafy.ai/help-center
• User install URL: https://capafy.ai/install-user-skill.md
• Publisher install URL: https://capafy.ai/install-publisher-skill.md