skills

Imported from luciousa/skills (AGENTS.md). Install upstream with npx skills add luciousa/skills. Copyright stays with the author.

Repository Guidelines

Project Structure & Module Organization

• skills/<skill-name>/SKILL.md contains portable, assistant-neutral workflows. Keep the workflow as the single source of truth.
• agents/*.agent.md are thin Copilot entry points that load a corresponding skill; do not duplicate the skill's methodology there.
• .github/instructions/ holds source specifications for the skills.
• skills/<skill-name>/evals/ contains synthetic prompts in evals.json and fixture repositories in files/.
• Generated evaluation output belongs in skills/<skill-name>-workspace/ or skills/<skill-name>-review/; both are ignored and must not be committed.

Development & Validation

This repository has no root package manifest, build script, or CI workflow. Validate Markdown changes by checking links, paths, front matter, heading structure, and consistency with the referenced files.

Run fixture tests from the fixture that owns them:

Set-Location skills/<skill-name>/evals/files/<fixture-name>
pytest

For focused Python checks, run the relevant pytest tests/<module>.py command from the fixture. Do not treat fixture test results as validation of the skills themselves; review generated output against each eval's expected output.

Writing Style & Naming

• Use Markdown with ATX headings, concise bullets, fenced examples, and repository-relative paths.
• Name skill directories in kebab-case, for example contextual-help-documentation; name agent files <role>.agent.md.
• Preserve YAML front matter keys (name, description) and keep descriptions specific.
• Ground claims in the active repository. Separate verified behavior from inference and explicitly state missing evidence instead of inventing commands, integrations, or runtime behavior.

Evaluation Guidance

Update evals.json when adding or changing a skill evaluation. Each evaluation needs a stable numeric id, clear prompt, expected output, and fixture path. Keep fixtures small and synthetic; add tests within the fixture when behavior needs evidence.

Commits & Pull Requests

The current branch has no commit history, so no established commit convention exists. Use concise imperative subjects such as Add contextual help skill eval. Keep each commit scoped to one skill, agent, or evaluation change.

Pull requests should summarize the change, list affected paths, explain validation performed, and include generated-output evidence when changing an evaluation. Link relevant issues when available. Do not commit local environments, caches, or generated workspaces.