Back to directory

pydantic-ai-skills

A companion library that adds remote skill registries and sandboxed execution of bundled files to Pydantic AI's Agent Skills support, for agents that load instructions and resources only when a task needs them.

pydantic-ai-skills

pydantic-ai-skills extends pydantic-ai-harness's support for Agent Skills — folders holding a SKILL.md description plus reference documents and scripts — with remote registries over Git and S3 (with filtering, prefixing, renaming and merging), execution of a skill's bundled reference documents and scripts through read_skill_resource and run_skill_script, sandboxed execution to keep untrusted scripts off the host, resolution of the ${SKILL_DIR} placeholder, and skills defined directly in Python.

It requires the harness package and delegates SKILL.md parsing, validation, the catalog and instruction rendering to it, so projects that only need plain instruction-based skills can use harness alone. Version 2 removed the older SkillsToolset and SkillsDirectory APIs in favor of harness's own load_capability.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.