Skill Seeker
Skill Seeker turns a documentation site, GitHub repo, PDF, or video into a packaged AI skill for Claude, Cursor, or another assistant in a few commands.
Install and create a skill
pip install skill-seekers && skill-seekers create https://docs.djangoproject.com/
About
Skill Seekers is a command-line pipeline that scrapes a source (a docs website, a GitHub repository with AST parsing to pull out real usage patterns, a PDF, a video transcript, a wiki export, or one of 18 supported source types) and turns it into structured, cited reference material an AI assistant can load as a skill. The common path is three commands: create a skill from a URL, then package it for whatever target you're deploying to.
Beyond scraping, it runs quality checks, including scoring gates and change detection against the original source, so a skill doesn't quietly drift out of date, and it can export the same extracted content to 22 different targets, including Claude, Gemini, OpenAI-style assistants, and LangChain. It also ships a Claude Code plugin with 35-40 MCP tools so the create-and-package workflow can run from inside an agent session instead of the CLI.
Key features
- Scrapes 18 source types: docs sites, GitHub repos, PDFs, videos, notebooks, wikis, and more
- AST-based parsing of code repos to extract real usage patterns and design conventions
- Packages output for 22 targets including Claude, Gemini, OpenAI, and LangChain
- Quality-scoring gates and change detection to keep a skill from drifting out of date
- Video transcript and frame extraction from YouTube, Vimeo, or local files
- Claude Code plugin exposing 35-40 MCP tools to run the pipeline from an agent session
Use cases
- Turning a framework's documentation site into a Claude skill in minutes
- Auto-detecting a project's stack and generating a starter skill config with the scan command
- Keeping a skill's content synced as the source documentation changes
- Building a skill once and exporting it for multiple AI assistants at once
Available tools
skill-seekers create <url>
Scrapes a documentation site (or other supported source) and builds a structured skill from it.
skill-seekers package <dir> --target claude
Packages an already-created skill for a specific target platform.
skill-seekers scan <project>
Auto-detects a local project's frameworks and generates a starter skill configuration.
MCP tool suite
35-40 MCP tools exposing the create/package/scan workflow to Claude Code, Cursor, Windsurf, and other MCP clients.
Frequently asked questions
Does it work on private or internal docs?
Yes. Anything reachable by the scraper (websites, repos you have access to, local PDFs or notebooks) can be a source; it isn't limited to public documentation.
Can I target more than one AI assistant from the same source?
Yes. Package the same extracted skill for any of the 22 supported targets without re-scraping.