Agent Work Skills

This repository publishes 16 independent skills for AI coding agents. It follows the open Agent Skills specification and works with skills.sh.

The catalog is intentionally modular. Install the skill that matches the task instead of loading the whole catalog by default. Each root SKILL.md is a short selector and workflow; detailed rules are in that skill's references/ directory and are read only when needed.

Catalog

Area	Skills
Working relationship	self, study, just-do-it
Coordinated work	cluster, point, line, plane
Product and developer experience	ux, nobody-cares-how, dx
Engineering judgment	simplicity, typesafe, effect, docs, pattern, testing

Choose by the question in front of you:

• point for one local cause and proof; line when one rule spans related modules; plane when one result crosses domains.
• study when an important unknown controls the next move; just-do-it to carry an accepted task through verification and handoff.
• cluster only when independent owners can make useful parallel progress.
• ux, nobody-cares-how, and dx for product, user-facing, and developer journeys.
• The engineering skills for the specific design, data, effect, documentation, pattern, or testing question.

Skills are explicit by default. A specialized skill should not be loaded merely because a task contains a related word. Use the normal $skill-name invocation when the skill's focused guidance is useful.

Structure

.
|-- AGENTS.md
|-- skills/
|   |-- <skill-name>/
|   |   |-- SKILL.md              # short selector and workflow
|   |   |-- references/            # detailed, task-specific guidance
|   |   |-- scripts/               # stable repeated operations
|   |   `-- agents/openai.yaml    # display metadata
|   `-- ...
|-- scripts/check-skill-layout
|-- devenv.nix
|-- pyproject.toml
`-- skills.sh.json

AGENTS.md contains the repository-local completion rules. It is deliberately contextual: agents should inspect the files relevant to the request, then implement, validate, repair, and finish without a mandatory full-repository reading ritual.

Install

Install one focused skill from the published repository:

npx skills add nmnmcc/skills --skill testing

Replace testing with any catalog name. To inspect all available skills without installing them:

npx skills add nmnmcc/skills --list

Install the complete catalog only when you are deliberately building a workflow that needs several independent skills:

npx skills add nmnmcc/skills

Validate

Install Devenv, then run the same checks used by CI:

devenv test

The check validates every skill with skills-ref, checks the compact-root and description budgets, and runs actionlint. The local fixtures and metadata are disposable and have no production access.

Writing rules

• Put each skill in skills/<skill-name>/SKILL.md with matching lowercase name frontmatter.
• Keep the root document a concise router: say when to use the skill, give the next useful steps, and link to deeper material.
• Keep description specific and short. The repository check caps it at 240 characters and the root document at 400 words.
• Put long guidance and ecosystem-specific material in references/; put stable repeated operations in scripts/.
• Keep agents/openai.yaml, skills.sh.json, README navigation, and the actual skill set in sync.
• Use direct, plain language. Do not commit secrets, tokens, personal data, or untrusted programs.

Release checks

• devenv test passes with the official skills-ref validator, the layout budget check, and actionlint.
• npx skills add . --list discovers all 16 catalog names.
• skills.sh.json is valid and lists each skill exactly once.
• Install examples use the published repository name.

License

MIT