/skill - Build or Enhance a Skill
Goal: Create a new skill or improve an existing one, guided by the skill-builder's design principles.
Skill Reference:
- skill-builder — Design principles, templates, and validation checklists
- skill-management — Governance rules for the skills library
When to Use
- After
/discoverflags a skill gap you want to fill - When a project needs a custom local skill for its domain patterns
- When you want to create a new global skill for the toolkit
- When an existing skill needs enhancement or audit
Steps
Step 1: Load Skill Builder
Read the skill-builder skill. The rest of this workflow follows its instructions.
Step 2: Gap Detection & Routing
Follow the skill-builder's Step 0: Gap Detection & Routing:
- Analyze the user's request — are they creating or modifying?
- Check
CATALOG.mdand.agents/skills/for existing matches. - Prior Art Check — before creating anything new:
- Does an MCP server already provide this capability?
- Does the vault have related notes or prior decisions (
/cx-search)? - Is there an existing workflow that already covers this use case?
- No Match → proceed to Step 3 (CREATE).
- Match Found → proceed to Step 5 (MODIFY).
Step 3: CREATE — Scope & Content Design
Follow the skill-builder's CREATE Mode:
- Scope Selection — Ask: global or local?
- Validation — Verify name and check for conflicts.
- Description Crafting — Write a trigger-rich description.
- Content Design — Define persona, anti-patterns, output shape, and relationships using the Design Principles.
Step 4: CREATE — Generate & Register
- Select Template — Basic, Intermediate, or Advanced based on content complexity.
- Generate Files — Scaffold the skill directory and
SKILL.md. - Register — Add to
CATALOG.md(global) or.agents/project-profile.md(local). - Verify — Run through both validation checklists (Structural + Design Quality).
Skip to Step 5.5.
Step 5: MODIFY — Audit & Enhance
Follow the skill-builder's MODIFY Mode:
- Full Audit — Evaluate against both Structural and Design Quality checklists.
- Gap Analysis — Categorize findings as Critical, Improvement, or Polish.
- Enhancement — Apply fixes in priority order (description → anti-patterns → progressive disclosure → examples → relationships).
- Update — Apply changes and update
CATALOG.mdif the description changed.
Step 5.5: Review & Score
This step is mandatory for both CREATE and MODIFY. Re-read the finished SKILL.md with fresh eyes and score it against this rubric. Each dimension is 0–2 points:
| Dimension | 0 | 1 | 2 |
|---|---|---|---|
| Description Quality | Generic or missing triggers | Has triggers but missing disambiguation or scenarios | Trigger-rich with clear "when to use" and "not for" signals |
| Persona & Purpose | Opens with mechanics ("This skill does X") | Has a persona but it's generic | Strong role framing that shapes the agent's mindset |
| Anti-Patterns | None | 1-2 generic warnings | 3+ specific, actionable "don't do this" items |
| Actionability | Theory only, no concrete examples | Has some examples but gaps | Concrete tables, templates, code blocks, or before/after examples |
| Progressive Disclosure | Everything jammed into SKILL.md (>8KB) or critical content in external files | Slightly over target but justified, or good structure with minor gaps | SKILL.md ≤5KB with heavy content properly extracted to docs/ |
Scoring:
- 9-10: Ship it. Exceptional quality.
- 7-8: Good. Minor polish possible but not blocking.
- 5-6: Needs work. Identify the weakest dimension and fix it before presenting.
- Below 5: Iterate. Re-apply the skill-builder's Design Principles and re-score.
Minimum threshold: 7/10. If the score is below 7, fix the weakest dimensions and re-score before proceeding. Present the score breakdown to the user in the summary.
Step 6: Summary
Present the result:
🛠️ Skill: <skill-name>
📍 Location: <path to SKILL.md>
📋 Mode: CREATE | MODIFY
🔖 Scope: Global | Local
✅ Structural Checklist: Passed
✅ Design Quality Checklist: Passed
📊 Quality Score: X/10
Description Quality: X/2
Persona & Purpose: X/2
Anti-Patterns: X/2
Actionability: X/2
Progressive Disclosure: X/2
Next: /commit to save changes
Usage
# Create a new skill (interactive)
/skill
# Enhance an existing skill
/skill enhance security-hardening
# Create a local skill for the current project
/skill create --local
Next Steps
After building or enhancing a skill:
- Use
/committo save changes - Use
/discoverto verify the skill appears in project recommendations