| name | create-skill |
| description | Create repository skills. Use for repeatable agent workflows. |
Create a skill
Create a skill only when a repeatable vertical workflow needs instructions beyond what code, tests, existing docs, and general model capability already provide.
Workflow
-
Collect concrete trigger and non-trigger examples.
-
Check existing skills for overlap; extend or consolidate before adding another metadata entry.
-
Choose one coherent job and a globally unique, verb-first kebab-case name.
-
Create .agents/skills/<name>/SKILL.md with only portable frontmatter:
---
name: <name>
description: <what it does and when to use it>
---
-
Put only essential procedure, project-specific gotchas, inputs/outputs, one compact example, and validation in SKILL.md.
-
Add a references/ file only for conditional detail and link it directly with an explicit read condition. Add a tested script only for repeated deterministic work.
-
Do not tell a skill to load a sibling skill. Give related skills precise, overlapping descriptions so hosts can compose them from the task.
-
Run pnpm link:aliases to refresh compatibility directory links.
-
Update AGENTS.md only when its compact routing index needs the skill.
-
Run pnpm check:workspace.
Read references/principles.md for scope/context decisions, references/structure.md for non-trivial resource layout, and references/patterns.md only when the workflow needs an established pattern. Do not load them by default.
Create another skill only when it has an independent trigger and useful standalone workflow. Keep supporting knowledge as a reference. Do not add auxiliary changelogs, duplicated project docs, generic tutorials, or variants that belong in one parameterized workflow.
Example
Input: “Create a repeatable workflow for reviewing package export changes.”
Output: One focused .agents/skills/review-package-exports/ skill with a precise trigger, compact procedure, example result, and validation loop.