| name | skill-creator |
| description | Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Claude's capabilities with specialized knowledge, workflows, or tool integrations. |
| license | Complete terms in LICENSE.txt |
Skill Creator
Create or update skills that add durable workflows, tool integrations, domain
knowledge, scripts, references, or assets without wasting context.
Load Detail On Demand
references/design-principles.md: token budget, degrees of freedom,
progressive disclosure.references/package-layout.md: SKILL.md, scripts, references, assets.references/creation-workflow.md: step-by-step creation/update process.references/workflows.md: workflow-writing patterns.references/output-patterns.md: output contracts and examples.
Use bundled scripts directly when useful:
scripts/init_skill.pyscripts/quick_validate.pyscripts/package_skill.py
Design Rules
- Keep
SKILL.md lean: trigger metadata plus essential workflow.
- Put detailed variants, schemas, long examples, and edge cases in
references/.
- Put deterministic or repetitive work in
scripts/.
- Put reusable output materials in
assets/.
- Do not add README, changelog, install guide, or process notes unless user asks.
- Avoid duplicating same guidance in
SKILL.md and references.
Workflow
- Clarify skill name, trigger use case, target agents, and expected outputs.
- Decide package shape:
- text-only workflow:
SKILL.md
- branching/deep guidance:
SKILL.md + references/
- fragile/repeated operation: add
scripts/
- reusable output material: add
assets/
- Write frontmatter
name and description so trigger conditions are clear.
- Write body as short decision tree plus required workflow.
- Move non-core examples/details into references and link them with "read when"
guidance.
- Validate structure and metadata.
Read only focused references needed for current skill design problem.
Validation
Run scripts/quick_validate.py <skill-dir> when available. Check:
- required
SKILL.md exists
- frontmatter has
name and description
- references/scripts/assets are linked from workflow when needed
- trigger text is specific enough to avoid accidental activation
- body fits progressive disclosure and avoids generic advice
