PiratePressPiratePress

PiratePress Skill for Claude Code, Kimi Code and Cursor

The PiratePress skill is an instruction pack that turns your AI agent into a video pipeline operator: the agent learns to write scripts, order videos through the API, review results and collect finished mp4s. Here is what's inside and how to use it.

What you need to prepare

  1. An AI agent with shell access — Claude Code, Kimi Code, Cursor, or a web agent with a terminal.
  2. An API key from the bot: @piratepress_bot → /apikey (shown once).
  3. A couple of minutes for installation.

How to install and what the skill does

Step 1. Install. One command via the skills CLI:

npx skills add piratepress/agents

Or with the one-liner curl -fsSL https://piratepress.fun/install.sh | PIRATEPRESS_API_KEY=pp_... bash — it installs both the skill and the MCP server, registering everything in the agents it finds automatically.

Step 2. Understand what the skill changes. It's not code — it's expertise: the agent gets a full map of the public API — endpoints, parameters, common combos, pricing, working modes. After installation, saying "make a video about …" is enough: the agent walks the flow itself — checks the balance, forms the order, waits for readiness, downloads the mp4 and the posting pack.

Step 3. Pick a working mode. The skill is self-sufficient: the whole REST flow inside runs on curl — suitable even for agents without MCP (like web agents with a shell). If the MCP server is installed, the skill prefers its tools (generate_video, wait_video, etc.) — the same API with less bookkeeping.

Step 4. Give the agent a real task. Scenarios from content-farm practice:

  • One-off video: "a 30-second EN clip about why houseplants die in winter" → order → mp4.
  • A batch: "5 videos on these topics, check the balance first" → the agent computes the cost, orders, collects all mp4s and posting metadata.
  • A scheduled pipeline: the agent generates a daily batch of videos from your RSS or notes and drops them into a publishing folder.
  • Quality review: in director mode the agent reads the script before rendering and edits it toward your style.

Common mistakes

A skill without a key. Instructions are there, funds are not: without PIRATEPRESS_API_KEY the agent hits a 401. The key is in the bot, via /apikey.

Vague orders in a pipeline. For series, ask the agent to use explicit parameters (generate_video) rather than a free-form prompt: predictable price and result.

No spend control. An agent with API access can spend dublons. Set the frame explicitly: "no more than N videos", "show me the estimate first" — the skill teaches the agent to check the balance.

FAQ

How does the skill differ from the MCP server? The skill is knowledge and flows (works anywhere with a shell). MCP is ready-made tools with progress and waiting. Together they're best; each is self-sufficient alone.

Which agents are supported? Any the skills CLI installs into: Claude Code, Kimi Code, Cursor and others — the installer auto-detects them.

Can I modify the skill? Yes — the piratepress/agents repository is open; fork it and add your own production rules.

Is it safe? The skill works with your key locally and sends nothing to third parties; the key is revoked by re-issuing it in the bot.

Can the skill run in CI/CD? Yes: the whole flow is curl against a REST API, so a pipeline assembles from plain steps: order → poll → download. Store the key in the runner's secrets.

What if the agent gets a parameter wrong? The API validates the order: conflicting parameters return a 422 with an explanation instead of burning dublons. The skill teaches the agent to read those responses and fix the order.

Does the skill work with custom-built agents? Yes: the instructions are platform-neutral. If your agent can read markdown instructions and call a shell — it can handle this.

Installation is one command: grab the key in @piratepress_bot and run npx skills add piratepress/agents — from there the agent makes videos itself.