Adopt Skill — Review & Integration Process

Evaluate an external skill before integrating it into this workflow. Never auto-adopt — always present findings and get user approval.

Step 1: Fetch & Read the Skill

1. Retrieve the skill from the provided source:
   • GitHub URL: Fetch the SKILL.md and any supporting files
   • Marketplace: Use /plugin to inspect before installing
   • Local path: Read the SKILL.md and directory contents
2. Read the complete SKILL.md content and all supporting files
3. Extract metadata: name, description, allowed-tools, context, dependencies

Step 2: Security Audit

4. Check for red flags:
   • Does it run arbitrary bash commands? What commands specifically?
   • Does it reference external URLs or APIs? Which ones?
   • Does it use allowed-tools that grant broad access?
   • Does it include scripts? Read them — do they do what they claim?
   • Does it instruct Claude to send data externally?
5. Rate security risk: Low / Medium / High
6. If High risk, stop and warn the user before proceeding

Step 3: Overlap Analysis

7. Read the skill catalog: .claude/skills/meta-skill-catalog/catalog.json
8. For each existing skill, compare:
   • Trigger overlap: Would both skills activate for similar prompts?
   • Functional overlap: Do they do similar things?
   • Naming conflict: Does the name collide with an existing skill?
9. Report findings:
   • "No overlap found" OR
   • "Overlaps with [skill-name]: [description of overlap]"
   • Recommend: adopt as-is, merge with existing, or skip

Step 4: Compatibility Check

10. Evaluate fit within this workflow:
    • Naming convention: Does it follow {category}-{function} pattern? If not, suggest a rename.
    • Category assignment: Which category does it belong to? (meta-*, dev-*, ops-*, tool-*, viz-*, or new category)
    • Context files: Does it need to read/write any context files? Does it respect the memory security rules (MEMORY.md main-session only)?
    • Quality: Is the SKILL.md well-structured? Are instructions clear?
    • Size: Is it under 500 lines? If not, should supporting files be split out?

Step 5: Adaptation Plan

11. Draft a plan for integration:
    • Proposed name (following naming convention)
    • Proposed category
    • Changes needed to make it fit:
      • Frontmatter adjustments (name, description, invocation control)
      • Instruction modifications for workflow compatibility
      • Supporting files to keep, modify, or remove
    • Any dependencies to install (MCP servers, CLI tools, packages)
12. If the skill needs modifications, show the diff between original and adapted version

Step 6: Present Review

13. Present a structured review to the user:

### Skill Adoption Review: {skill-name}

**Source:** {URL or path}
**Security Risk:** {Low/Medium/High}
**Overlap:** {None / overlaps with X}
**Category:** {proposed category}
**Proposed Name:** {category-function}

#### What It Does
{One paragraph summary}

#### Changes Needed
- {List of modifications}

#### Recommendation
{Adopt as-is / Adopt with changes / Merge with [existing] / Skip}

14. Wait for user approval before proceeding

Step 7: Install (After Approval)

15. If approved:
    • Copy/create the skill directory: .claude/skills/{category-function}/
    • Write the adapted SKILL.md (with any agreed modifications)
    • Copy supporting files
    • Update the skill catalog: .claude/skills/meta-skill-catalog/catalog.json
    • Set version tracking in catalog entry:
      • version: read from the adopted skill's SKILL.md frontmatter (or 1.0.0 if not present)
      • installed_from: "local" (adopted skills are local additions)
      • installed_version: same as version
      • installed_date: today's date
    • Update today's daily memory with the adoption decision
16. Verify the skill is discoverable: "What skills are available?"
17. If the skill has a test case, run it to verify it works

Adoption Checklist

Before marking adoption complete, verify:

• SKILL.md has proper frontmatter (name, description)
• Name follows {category}-{function} convention
• No security red flags unaddressed
• No unresolved overlap with existing skills
• Catalog updated
• Supporting files in place
• Skill loads when invoked