Skill Library Cleaner Normalize Canonicalize

Skill Library Cleaner / Normalize & Canonicalize

Role

Turn dedupe decisions into a clean target library plan.

Input:

machine/DEDUP_GROUPS.json
machine/RISK_BLOCKLIST.json

Output:

reports/CANONICAL_LIBRARY_PLAN.md
machine/CANONICAL_LIBRARY_PLAN.json

Required Decisions

For each group, decide:

• canonical name;
• canonical folder name;
• category;
• short description;
• target path;
• source/provenance note;
• keep/archive/block/reference/backlog route;
• whether a SKILL.md rewrite is allowed;
• what evidence supports the rewrite.
• rewrite risk: none, minor, meaning-change, or blocked.
• confidence and manual-review reason when confidence is not high.

Naming Rules

• Use clear lowercase kebab-case folder names.
• Avoid private product names unless the skill is private-only.
• Avoid copied third-party wording.
• Keep one strong canonical default instead of many near-duplicates.
• Preserve source/provenance notes.
• Prefer role-based names over brand/source names for generic public skills.
• Preserve private/user-owned naming when the skill is intentionally local.
• Do not rewrite a SKILL.md when the cleaner cannot preserve the skill's trigger, safety boundary, and output contract.

Shelf Rules

Target routes:

• canonical: active skill the user can use.
• references: useful idea/source, not an active skill.
• backlog: promising but incomplete or needs rewrite.
• blocked: risky until approved or rewritten.
• archive: exact or approved duplicate, never a permanent delete.

For broad overlap clusters, produce a shelf plan instead of pretending they are exact duplicates.

Pass Condition

The apply step can create a clean library without guessing names, categories, or destinations.