conda-skills

conda skill makes skills from conda packages available to your agent through directory links in a project or your home directory. Conda manages the package versions. You choose which skills to use.

Inspired by Pavel Zwerschke's pixi-skills and skill-forge, conda-skills brings the same package-manager approach to conda's Python stack. See prior art for the background and shared conventions.

Requires Python 3.11 or newer and conda 26.9 or newer. Install conda through a conda distribution, not PyPI.

Installation

Install a published package into the environment that runs conda, usually base:

conda install -n base --override-channels -c jezdez -c conda-forge conda-skills
conda skill --help

To use a checkout, install it with conda-pypi in the environment that runs conda:

conda install -n base --override-channels -c conda-forge "conda-pypi>=0.12" "hatchling>=1.27" "hatch-vcs>=0.4" "rich>=13" "ruamel.yaml>=0.11.14,<0.20"
conda pypi install -n base -e .
conda skill --help

Run these commands from the repository root. The first supplies build requirements and runtime dependencies, which editable installs do not resolve. If another prefix owns your conda executable, replace -n base with -p /path/to/conda-prefix in both commands.

The first-use tutorial creates and activates a new skill-demo environment, then uses conda skill list and conda skill link to review and connect a skill to a project. Install the plugin before activating that environment. Conda's initialized shell keeps using the conda executable that owns the plugin.

See installation for PyPI and local-artifact alternatives. Installing the plugin in an unrelated environment does not register it with another conda executable.

Documentation

• Documentation home
• Install conda-skills
• Try your first skill
• Share skills with conda-workspaces
• Update or roll back a skill
• Build and try a skill package
• Watch the demos
• Package a skill
• Commands and JSON output
• How agents find skills
• Skill safety
• Troubleshooting

The default destination is .agents/skills/ under the current directory. Use --backend claude for .claude/skills/ and --global for your home directory. Read skill content before exposing it to an agent. The plugin does not execute instructions or verify their trustworthiness.

conda skill unlink NAME removes the destination link independently of the active environment, leaving its package installed. It verifies the link's target against conda package records. For a broken or unverifiable target, inspect conda skill status before using --force. That option removes only a symlink and never bypasses redirected-parent checks. Explicit -n/-p source overrides for list and link are covered in Use an installed skill.

Development

Development environments and tasks live in the root conda.toml, with resolved dependencies in conda.lock. Install conda-workspaces, then install each environment before running its tasks:

conda install -n base --yes --override-channels -c conda-forge 'conda>=26.9' 'conda-workspaces>=0.12'
conda workspace install --locked -e default
conda task run -e default check
conda workspace install --locked -e docs
conda task run -e docs docs

See Contributing, Releasing conda-skills, and the Changelog. The Docs workflow checks pull requests and deploys documentation from main to GitHub Pages.

License

Original code is licensed under Apache-2.0.