| name | skill-creator |
| description | Use when users want to create a new skill or improve an existing one, especially when they need help defining scope, choosing a skill pattern, designing activation boundaries, structuring references or scripts, or validating the final package. |
| metadata | {"internal":true} |
Skill Creator
Overview
Design first, implementation second. The goal is a minimal, executable skill whose trigger boundary is clear and whose runtime instructions stay separate from design, testing, and packaging notes.
Workflow
- Define scope before editing files.
- Capture the task, trigger prompts, success criteria, non-goals, and required tools or MCPs.
- If the skill is new or underspecified, read
references/workflows.md.
- Choose the dominant pattern before drafting
SKILL.md.
- Read
references/pattern-selection.md.
- Use one primary pattern unless a second pattern is necessary to prevent a concrete failure mode.
- Design activation before implementation.
- Read
references/activation-design.md.
- Write trigger prompts, non-trigger prompts, false-positive risks, false-negative risks, and the first action after activation.
- Split files by responsibility.
- Keep runtime protocol in
SKILL.md.
- Put detailed domain rules, checklists, and examples in
references/.
- Put copied or consumed output resources in
assets/.
- Add
scripts/ only when a manual step is repetitive or needs deterministic execution.
- Write the minimal runtime protocol.
- Frontmatter: only
name and description.
- Description: explain when to use the skill, not how it works.
- Body: include workflow, guardrails, error handling, and output expectations.
- Validate and package deliberately.
- Read
references/validation-and-distribution.md.
- Add trigger, functional, and regression prompts as internal test assets when the skill will be maintained.
- Package only the runtime files that should ship.
Guardrails
- Do not expand
SKILL.md before the scope, pattern, and activation boundary are clear.
- Do not keep long design notes, activation analysis, or test methodology in
SKILL.md.
- Do not scaffold placeholder files "just in case". Add
scripts/ and assets/ only when they solve a real repeated need.
- When validation fails, fix the right layer:
- trigger issue -> activation design or description
- workflow issue ->
SKILL.md
- missing domain rule ->
references/
- brittle manual step ->
scripts/
- package bloat -> distribution selection
Reference Map
references/workflows.md: recommended sequence for building a skill from scope to distribution
references/pattern-selection.md: how to choose among Tool Wrapper, Generator, Reviewer, Inversion, and Pipeline
references/activation-design.md: trigger boundary checklist and description rules
references/validation-and-distribution.md: test assets, verification flow, and runtime packaging rules
Commands
python3 scripts/init_skill.py my-skill --path /path/to/skills
python3 scripts/quick_validate.py /path/to/skills/my-skill
python3 scripts/package_skill.py /path/to/skills/my-skill ./dist