Plugin Optimization

Execute plugin validation and optimization workflow. Target: $ARGUMENTS

Background Knowledge

Tool Invocation Patterns

Qualified Names: Plugin components MUST use plugin-name:component-name format.

Progressive Disclosure: Skills package domain expertise—turning general agents into knowledgeable specialists. Progressive disclosure protects context: metadata (~50 tokens) -> SKILL.md (~500 tokens) -> references (2000+ tokens, MUST only access when specifically needed).

Component Template Validation

Components MUST conform to templates in ${CLAUDE_PLUGIN_ROOT}/examples/.

Agent Template Requirements:

• Frontmatter: name, description, color, allowed-tools
• System Prompt: Second person (“You are…”), descriptive voice (NOT directive)
• Structure: Knowledge Base + Core Responsibilities + Approach sections

Instruction-Type Skill Requirements (user-invocable: true):

• Frontmatter: name, description, user-invocable: true, allowed-tools
• Voice: Imperative (“Load…”, “Create…”, “Execute…”)
• MANDATORY Phase Format: ## Phase N: [Phase Name] with Goal and Actions subsections
• Optional Pre-Phases: ## Background Knowledge, ## Initialization, ## Context

Knowledge-Type Skill Requirements (user-invocable: false):

• Frontmatter: name, description, user-invocable: false
• Voice: Declarative (“is”, “are”, “provides”)
• Structure: Topic-based sections (“## Core Concepts”, “## Best Practices”)

Type Classification Quick Reference:

Manifest Cross-Validation: Instruction-type MUST be in plugin.json commands array; Knowledge-type MUST be in plugin.json skills array.

Phase 1: Discovery & Validation

Goal: Validate structure and detect issues. Orchestrator MUST NOT apply fixes.

Actions:

1. Resolve path with realpath and verify existence
2. Validate .claude-plugin/plugin.json exists
3. Find component directories: commands/, agents/, skills/, hooks/
4. Validate components against ${CLAUDE_PLUGIN_ROOT}/examples/ templates
5. Assess architecture: if commands/ exists with .md files, use AskUserQuestion tool to ask about migrating to skills structure
6. Run validation: python3 ${CLAUDE_PLUGIN_ROOT}/scripts/validate-plugin.py "$TARGET"
   • Options: --check=structure,manifest,frontmatter,tools,tokens
   • JSON output: --json
   • Verbose: -v, --verbose
7. Compile issues by severity (Critical, Warning, Info)

Phase 2: Agent-Based Optimization

Goal: Launch agent to apply ALL fixes. Orchestrator does NOT make fixes directly.

Condition: Always execute.

Actions:

1. Launch plugin-optimizer:plugin-optimizer agent with the following prompt content:
   • Target plugin path (absolute path from Phase 1)
   • Validation console output (issues list from Phase 1)
   • Template validation results
   • User decisions (migration choice if applicable)
   • INSTRUCTION: Analyze the validation output to identify issues
2. Agent autonomously applies fixes (MUST use AskUserQuestion tool before applying template fixes, presenting violations with specific examples and before/after comparison)
3. Agent increments version in .claude-plugin/plugin.json after fixes:
   • Patch (x.y.Z+1): Bug fixes
   • Minor (x.Y+1.0): New components
   • Major (X+1.0.0): Breaking changes
4. Wait for agent to complete

Path Reference Rules:

• Same directory: Use relative paths (./reference.md)
• Outside directory: Use ${CLAUDE_PLUGIN_ROOT} paths
• Component templates: See ${CLAUDE_PLUGIN_ROOT}/examples/

Redundancy & Efficiency:

• Redundancy: Allow strategic repetition of critical content (MUST/SHOULD requirements). Favor concise restatement.
• Efficiency: Agent detects if tasks need Agent Teams (Parallelizable > 5 files, Multi-domain).

Phase 3: Verification & Deliverables

Goal: Verify fixes, generate report, and update documentation.

Actions:

1. Execute validation script: python3 ${CLAUDE_PLUGIN_ROOT}/scripts/validate-plugin.py "$TARGET"
2. Analyze results: compare with Phase 1 findings, confirm critical issues resolved
3. If critical issues remain, resume agent execution
4. Generate final validation report using template below
5. Update README.md to reflect current state (metadata, directory structure, usage instructions; do not append version history log)

Validation Report Template

## Plugin Validation Report

### Plugin: [name]
Location: [absolute-path]
Version: [old] -> [new]

### Summary
[2-3 sentences with key statistics]

### Phase 1: Issues Detected
#### Critical ([count])
- `file/path` - [Issue description]

#### Warnings ([count])
- `file/path` - [Issue description]

### Phase 2: Fixes Applied
#### Structure Fixes
- [Fix description]

#### Template Conformance
- **Agents**: [Count] validated, [count] fixed
- **Instruction-type Skills**: [Count] validated, [count] fixed
- **Knowledge-type Skills**: [Count] validated, [count] fixed

#### Redundancy Fixes
- [Consolidations applied]

### Phase 3: Verification Results
- Structure validation: [PASS/FAIL]
- Manifest validation: [PASS/FAIL]
- Component validation: [PASS/FAIL]
- Tool patterns validation: [PASS/FAIL]
- Token budgets validation: [PASS/FAIL]

### Token Budget Analysis
- Skills analyzed: [count]
- Tier 1 (Metadata ~50): [OK count], [WARNING count]
- Tier 2 (SKILL.md ~500): [OK count], [WARNING count], [CRITICAL count]
- Tier 3 (References 2000+ typical): [total tokens]

### Component Inventory
- Commands: [count] found, [count] valid
- Agents: [count] found, [count] valid
- Skills: [count] found, [count] valid

### Remaining Issues
[Issues that couldn't be auto-fixed with explanations]

### Overall Assessment
[PASS/FAIL] - [Detailed reasoning]