Skip to main content
The ContentHero skill is an installable set of workflows for your agent. It teaches your agent to use ContentHero well: how to ground a draft, how to chain a generation, how to prompt each model, how to assemble and publish. It runs inside your AI (Claude, ChatGPT, Gemini, Codex, Cursor, and more) and drives the MCP, CLI, or raw API underneath. It ships as a plugin that brings the ContentHero connector with it: see Install the Skill. The repo is contenthero-ai/skills.

​
What it covers

Everything on the public surface. The skill routes your request to the right workflow, so there is nothing to choose before you start.

Generate

Image, video, audio, reference boards, lip-sync, upscaling. Cost preflight before every spend, and output-id chaining so you never re-upload.

Plan and publish

Spaces, stages and cards, one post per platform, scheduling, and publishing to your connected accounts.

Ground and draft

Your brand kit and its searchable knowledge base. Your LLM writes in your voice; you approve before anything goes live.

Research

The accounts you track and their posts ranked by outlier score, which measures a post against its own creator’s baseline rather than the platform’s.

Media library

Upload, import, search, folders, and archiving that is always reversible.

Editor

Projects, timelines, canvases, saved elements, renders of your work, and exports.

​
What the skill adds

ContentHero supplies the tools. The skill supplies the craft of using them well, which is the part a tool schema cannot carry:
  • The order to do things in, so the agent grounds before it drafts and prices before it spends.
  • The traps that cost money or delete work. A card’s posts and assets are declarative, so a partial update removes what you left out. A slow render is not a failed one, and re-submitting charges you twice.
  • Model prompt craft specific to our roster, so a prompt is tuned to the model that will run it.
  • Grounded voice synthesis, the method for drafting copy that sounds like you (see Voice synthesis).
It deliberately does not restate the tool reference. Your agent already receives every tool’s name, description and full schema when it connects, so repeating them would cost context and drift out of date. Parameter detail lives in the MCP tool reference, generated from the published packages.

​
Coverage is enforced, not claimed

The skill’s CI reads the live tool surface from the published @contenthero/mcp and checks it in both directions: every tool the skill names must exist, and every tool that exists must be named in a workflow. A build fails if either drifts. The second direction is the one that matters. A skill can claim it covers a product and be wrong by most of the surface, and no test, build or typecheck would ever say so. This one cannot ship in that state.

​
The frozen-snapshot model

An installed skill is a snapshot. It does not auto-update; you update it with git pull or by re-running the installer. The skill is only instructions. The services it drives (the hosted MCP, the CLI, the model roster, pricing) are server-side and always live. So the skill bakes in durable guidance (craft, sequence, traps) and defers volatile facts (model ids, prices, platform limits) to live calls. That is what keeps a year-old install correct. This is also why MCP ranks first in the auth ladder: it is the only self-updating transport.

​
Install

Install the skill

Per-host install, the paste-prompt one-liner, and how to authenticate.