auto-skill-generator

The Auto Skill Generator is a specialized engineering utility designed for developers and AI agents to transform scattered web documentation into structured, reusable agent skills. By automating the research and crawling process, this tool allows users to quickly build a library of skills for specific libraries, APIs, or frameworks like Stripe, Prisma, or Modal. It bridges the gap between raw documentation and functional agent knowledge by enforcing a consistent structure that includes quick-start guides, core API references, and common coding patterns. It is ideal for developers who need to rapidly onboard agents to new technical stacks or build comprehensive automation workflows from authoritative web sources.

• Automated documentation research using intelligent search queries for official docs, readthedocs, and guides.

• Precision crawling capabilities with domain filtering, path inclusion/exclusion regex, and custom instruction support to focus on relevant API methods.

• Structured skill generation that produces standard-compliant .md files ready for immediate integration into the SkillsBench framework or agent environments.

• Configurable depth and extraction quality to optimize for token usage and data relevance, ensuring high-quality code block extraction.

• Domain-specific sanitization to ignore marketing content, changelogs, and irrelevant blog posts to maintain a clean knowledge base.

• Use this skill when prompted with requests like "create a skill for [topic]" or "generate a guide for [API]."

• Always target official authoritative sources like *.com/docs/ or *.readthedocs.io to ensure accuracy.

• The generated skills should be saved into the standard ~/.claude/skills/ directory structure for immediate accessibility by the agent.

• Keep the generated SKILL.md under 500 lines for optimal performance; if documentation is extensive, offload supplementary information to a references/ subdirectory.

• Leverage the --no-external flag during crawling to prevent the crawler from straying into off-topic websites or external marketing domains.