A Claude Code skill that reads your codebase and writes a complete help center — no templates, no manual work. It analyzes routes, components, config schemas, and error messages to build a content plan, then writes professional articles optimized for AI agent retrieval. Every claim traces back to actual source code. Articles are structured so tools like Intercom’s Fin, Zendesk AI, or any RAG system can retrieve and use them accurately. Plain markdown output — import anywhere, no vendor lock-in.
I built this to solve a specific problem: you’re a startup, you want to set up an AI support agent, but it needs a knowledge base — and you don’t have one. Writing help articles from scratch takes weeks, and you’ve got a hundred other things to ship.
But the answers already exist in your codebase. Every feature, config option, and error message is already defined in code. The missing step was extracting that into articles structured for how AI agents actually retrieve and use information.
Help Content from Code is a Claude Code skill that does exactly that. Point it at your project and it generates a full set of help articles — grounded in source code, optimized for RAG retrieval, ready to import into Fin or any other customer support AI Agent.
Some examples from the demo site:
- Ollama → 13 articles, 18,637 words
- QMD → 8 articles, 16,064 words
- Pearcleaner → 12 articles, 17,358 words
Install with `npx skills add gustavscirulis/help-content-from-code`, open any repo, and ask it to write help articles. Also works with GitHub Copilot, Cursor, and Gemini CLI.
Would love to see what you generate with it!
Fin