kolega-skills

Kolega Skills

A public collection of Agent Skills maintained by Kolega. Skills are self-contained folders of instructions and optional resources that give an AI agent reliable workflows for specialized tasks.

This repository follows the open Agent Skills specification and adds repository-level validation, deterministic packaging, tests, and CI.

Catalog

Every directory under skills/ is an independently installable skill. A skill contains a required SKILL.md and may include:

skill-name/
├── SKILL.md       # Required metadata and instructions
├── scripts/       # Optional deterministic utilities
├── references/    # Optional documentation loaded as needed
└── assets/        # Optional templates and static resources

Use a skill

Install a complete directory from skills/ through any Agent Skills-compatible client's skill-management flow, or copy it into a project or user skill directory configured by that client. Keep the directory intact: resources are referenced relative to the skill root.

Once installed, describe the task normally. A compatible client uses each skill's name and description to decide when to load its instructions.

Document skill runtimes

Each document skill is independently installable and carries its own direct Python runtime requirements with compatible release ranges. The skill selects any available Python 3.11+ interpreter rather than assuming that python or python3 has a particular meaning. Before installing anything, it tells the user what is missing, where it will be installed, and which installer it intends to use. Python packages are normally installed through the selected interpreter's -m pip. Python itself and external applications use the platform's normal package manager, with Homebrew preferred on macOS when available. A local environment is a fallback when direct installation is blocked; it is not required. The document skills do not require uv; uv is used only by the repository development and validation commands below.

• DOCX-to-PDF and PPTX-to-PDF conversion optionally use a separately installed LibreOffice soffice executable. The source-format skill owns these conversions; use pdf for operations on an existing PDF.
• PDF OCR is optional and never runs automatically. Use ocr-plan first, then explicitly choose an independently installed engine and locally provisioned model or language data: Surya for modern layout-aware extraction when its model terms and runtime are suitable, PaddleOCR as the no-GPU fallback, or Tesseract as an explicit last-resort fallback when the preferred engines are unavailable. Tesseract is best suited to clean printed text on CPU when flat output is acceptable.
• No OCR engine, model weights, language data, or LibreOffice binary is bundled. Optional runtimes are installed only after the skill tells the user about the intended system change. OCR models and language data remain explicit, reviewed local artifacts and are never downloaded implicitly.
• Surya's code is Apache-2.0, but its current model weights use modified AI Pubs Open Rail-M terms with commercial-use eligibility limits. Review and approve the selected model's terms before provisioning it. Verify model and language-data licenses for every OCR engine.

Create a skill

Requirements:

• uv

Sync the repository tooling environment:

uv sync --locked

Create a skill from the repository template:

uv run python scripts/new_skill.py my-skill \
  --description "Describe what the skill does and when Kolega should use it." \
  --resource references \
  --resource scripts

Only request resource directories the skill actually needs. Then replace the generated instructions with a complete workflow and add focused resources.

Validate and package

Validate every skill and run the tooling tests:

uv run python scripts/validate_skills.py
uv run python -m unittest discover -s tests -v
uv run ruff check .
uv run ruff format --check .

Build deterministic .skill archives with SHA-256 sidecars:

# Package every skill into dist/
uv run python scripts/package_skill.py --all

# Package one skill elsewhere
uv run python scripts/package_skill.py skills/skill-authoring --output /tmp/kolega-skills

A .skill file is a ZIP archive whose top-level directory is the skill name. Packaging first runs the same validation used in CI.

Contribute

Read CONTRIBUTING.md before submitting a skill. Contributions must be original or redistributable, self-contained, narrowly scoped, and free of secrets. CI validates metadata, local links, file safety, tests, and package reproducibility.

License

Unless a skill states otherwise, this repository is licensed under the Apache License 2.0. Third-party content must include its own license and attribution inside the skill.