Chezmoi agent skills

Agent skills for use with chezmoi.

These skills follow the agent skills specification and use progressive disclosure so agents load detailed references only when needed.

Install

Each directory under skills/ is a portable skill source with a SKILL.md entry point.

Add the marketplace with claude plugin marketplace add mkobit/chezmoi-skills.

Install the plugin with claude plugin install chezmoi@chezmoi-skills.

Add the marketplace with codex plugin marketplace add mkobit/chezmoi-skills.

Launch codex, run /plugins, and install Chezmoi from the marketplace UI.

Copy or symlink the required directories from skills/ into the host's documented skill-discovery location.

Cursor discovers project skills from .agents/skills/ and .cursor/skills/, including nested directories.

AGY and other Agent Skills consumers require their own documented discovery or packaging steps.

The repository contains a Claude marketplace manifest at .claude-plugin/marketplace.json, a portable Agent Plugins manifest, and a Codex compatibility manifest at .codex-plugin/plugin.json.

Host	Current status
Claude Code	The repository supplies a Claude marketplace manifest, but installation has not been exercised in automated integration tests.
Codex	The portable and Codex compatibility manifests are present, but this package has not yet passed an end-to-end Codex installation test.
Cursor	The skill layout now satisfies Cursor's documented `name`-to-directory requirement, but this package has not yet passed an end-to-end Cursor installation test.
AGY	Compatibility has not been assessed.
Other Agent Skills consumers	The skills follow the open specification, but each host's discovery, packaging, and frontmatter requirements must be verified before claiming support.

Do not treat the table as a support guarantee.

Install the pinned Bun version and project tools with mise install.

Install locked dependencies with bun install --frozen-lockfile.

Run the structural checks with mise run validate.

Run deterministic contract tests with mise run test:contracts.

Run unit tests with mise run test.

Run all linters with mise run lint.

Generate and validate the Promptfoo suites with bun run eval:validate; this path does not call providers.

Credentialed evaluations are sparse, manual, and reviewed, and they are not routine continuous-integration or release checks.

See the evaluation strategy and live benchmark review for longitudinal result tracking, OpenAI smoke samples, Codex host checks, and full-matrix gates.

Keep skill entry points concise and move detailed material into linked references/ files.

Use conventional commits and semantic pull-request titles.

Use feat or fix for skill-content changes that should produce a release, and use docs only for repository documentation.

Open a pull request against main, wait for required checks, and merge through the pull-request workflow.

Release Please creates releases from eligible commits on main and synchronizes manifest versions.

This project is licensed under the MIT License.