Plugin Optimization

Analyze and optimize a plugin's knowledge and skills for size and clarity.

Parse Arguments

Parse $ARGUMENTS:

• First positional (optional): Path to plugin directory. Default: current working directory.
• --dry-run: Report changes without applying. This is the default — changes are only applied when user explicitly omits this flag.
• --target: Focus area — knowledge (default), skills, or all.

Cache guard: ~/.claude/plugins/cache/ is READ-ONLY. Resolve to the real git repo via atlas: atlas_search_projects(query="plugin-name").

Knowledge Optimization (target: knowledge | all)

• Measure each file, flag > 10 KB or total > 50 KB
• Find duplicates, generic content, outdated refs, verbose sections
• Apply: remove duplicates (keep detailed version), condense, remove dead refs, split oversized files

Skills Optimization (target: skills | all)

For each skill, evaluate:

Description quality (most impactful — poor descriptions mean skills never trigger):

• Does it include specific trigger phrases a user would say?
• Does it have proactive language ("Use when...", "Trigger after...") if appropriate?
• Could another plugin's skill "steal" the trigger due to overlapping descriptions?
• Is it too vague ("manages things") or too narrow (only one exact phrase)?

Instruction clarity:

• Are steps concrete enough for an agent to follow without guessing?
• Are knowledge file references valid (files exist)?
• Are there stale references to removed features or files?
• Is there content drift between paired skill/agent files?

Cross-skill consistency:

• Consistent argument parsing patterns across skills?
• Cache guard reminder present where editing happens?
• Output format defined clearly?

Apply: fix broken refs, sharpen descriptions, align skill/agent pairs, clarify steps.

For deeper skill refinement (test prompts, eval loops, description optimization), recommend /skill-creator (from the plugin-dev plugin).

Post-Optimization

Prepend optimization entry to REFLECTIONS.md (format from knowledge/lifecycle-formats.md)

Output

Optimization {applied|dry-run} for {plugin-name} v{version}

Knowledge: {before} KB → {after} KB ({reduction}%)
  - {file}: {change}

Skills: {N} modified
  - {skill}: {change}