dais-skills

A Python package for programmatically discovering, downloading, and parsing Agent Skills.

Currently supports GitHub as a skill source. GitLab, arbitrary Git repositories, and custom sources (e.g. S3) are planned.

This project is in early development. The API may change at any time and backward compatibility is not guaranteed yet.

中文文档

Features

• Scan a repository: given a GitHub repo URL, list every skill in it along with its path, name, and description
• Download a skill: given a repo URL and a skill path, package that skill directory as a zip byte stream
• Parse a skill archive: extract SKILL.md metadata (with frontmatter) and every resource file (auto-classified as text or binary) from a zip

Installation

Requires Python 3.14+.

uv add dais-skills
# or
pip install dais-skills

Quick Start

Scan a repository for skills

import asyncio
from dais_skills import scan_repo

async def main():
    skills = await scan_repo("https://github.com/<owner>/<repo>")
    for s in skills:
        print(s.path, "-", s.name)
        print(" ", s.description)

asyncio.run(main())

scan_repo returns list[ScannedSkill], where each item has:

• path: the skill directory path inside the repo
• name: the name field from the SKILL.md frontmatter
• description: the description field from the SKILL.md frontmatter

The scanner walks the priority directories defined by the Agent Skills specification (such as skills/, .claude/skills/, .github/skills/, and so on) and skips irrelevant directories like node_modules, .git, dist, build, and __pycache__.

Download a single skill as a zip

import asyncio
from dais_skills import download_skill_zip

async def main():
    data: bytes = await download_skill_zip(
        "https://github.com/<owner>/<repo>",
        "skills/my-skill",
    )
    with open("my-skill.zip", "wb") as f:
        f.write(data)

asyncio.run(main())

skill_path may point at either the skill directory or the SKILL.md inside it. Both forms are treated as equivalent.

You can also pass an existing httpx.AsyncClient via the optional keyword-only argument client to reuse connections across calls.

Parse a downloaded skill archive

import zipfile
from dais_skills import Skill

with zipfile.ZipFile("my-skill.zip") as zf:
    skill = Skill.from_zip(zf)

print(skill.name, skill.description)
print("license:", skill.license)
print("compatibility:", skill.compatibility)
print("allowed-tools:", skill.allowed_tools)
print("metadata:", skill.metadata)
print("body:", skill.content[:80])

for res in skill.resources:
    if res.type == "text":
        print("[text]", res.relative, len(res.content))
    else:
        print("[bin] ", res.relative, len(res.content))

Skill.from_zip will:

1. Locate the skill root inside the zip (the first-level directory containing SKILL.md, or the archive root)
2. Parse the frontmatter and body of SKILL.md
3. Collect every remaining file as a SkillResource, splitting them into TextResource and BinaryResource via binaryornot

Public API

Exported from the top-level package:

GitHub client exceptions (GitHubError, InvalidGitHubUrlError, GitHubApiError, GitHubTreeFetchError, GitHubBlobFetchError) are available from dais_skills.public.

Development

uv sync
uv run pytest              # run unit tests
uv run pytest -m smoke     # run end-to-end smoke tests that hit real network services

Roadmap

• Scan skills in a GitHub repository
• Download and package a single skill from GitHub
• Parse a skill archive
• GitLab repository support
• Arbitrary remote Git repository support
• S3 and other custom sources