Marketplace Optimizer

Orchestrates specialized sub-agents to optimize Claude Code marketplaces efficiently. Each agent loads only the documentation it needs, keeping context usage minimal.

Architecture

marketplace-optimizer (this skill)
    │
    ├── optimizer-skills     → skills.md, github-skills-readme.md
    ├── optimizer-commands   → slash-commands.md
    ├── optimizer-agents     → sub-agents.md
    ├── optimizer-hooks      → hooks.md
    ├── optimizer-mcp        → mcp.md
    │
    └── optimizer-marketplace → plugins.md, plugins-reference.md,
                                github-*.md, CHANGELOG

When This Skill Activates

• User wants to optimize the marketplace or plugins
• User wants to optimize project-level .claude/ elements
• User asks to sync with latest Claude Code best practices
• User mentions "optimize", "update docs", or "best practices"

Scope

Project-level (.claude/):

• Skills, Agents, Commands

Marketplace (plugins/):

• Skills, Commands, Agents, Hooks, MCP configs

Execution Protocol

Phase 1: Cache Update

Step 1: Version Check

Flags:

• --update-cache: Force update, skip check
• --skip-cache: Skip update entirely

Without flags:

bun .claude/scripts/check-cache-version.ts

Step 2: Update Cache (if needed)

bun .claude/scripts/update-docs-cache.ts

Phase 2: Secondary Sources (Optional)

If no arguments provided, prompt:

Sekundäre Quellen (optional)

Zusätzliche Referenzen eingeben (URLs oder lokale Dateien), oder leer lassen:

• Empty / "keine" / "none" / "no": Skip secondary sources
• Otherwise: Parse as URLs and/or local file paths

Phase 3: Parallel Analysis

Launch specialized agents in parallel using the Task tool:

## Agents to Spawn

| Agent | Scope | Docs Loaded |
|-------|-------|-------------|
| `optimizer-skills` | All skills in .claude/ and plugins/ | skills.md, github-skills-readme.md |
| `optimizer-commands` | All commands | slash-commands.md |
| `optimizer-agents` | All agents | sub-agents.md |
| `optimizer-hooks` | All hooks.json + scripts | hooks.md |
| `optimizer-mcp` | All .mcp.json | mcp.md |

Spawn all 5 agents in parallel:

Task(subagent_type="optimizer-skills", prompt="Analyze all skills in .claude/skills/ and plugins/*/skills/. Read docs first, then check each SKILL.md.")

Task(subagent_type="optimizer-commands", prompt="Analyze all commands in .claude/commands/ and plugins/*/commands/. Read docs first, then check each command.")

Task(subagent_type="optimizer-agents", prompt="Analyze all agents in .claude/agents/ and plugins/*/agents/. Read docs first, then check each agent.")

Task(subagent_type="optimizer-hooks", prompt="Analyze all hooks in plugins/*/hooks/. Read docs first, then check hooks.json and scripts.")

Task(subagent_type="optimizer-mcp", prompt="Analyze all MCP configs in plugins/*/.mcp.json. Read docs first, then check each config.")

Phase 4: Cross-Reference & Feature Analysis

After parallel agents complete, spawn the marketplace agent:

Task(subagent_type="optimizer-marketplace", prompt="
Analyze overall marketplace structure:
1. Read plugins.md, plugins-reference.md, github-*.md, github-changelog.md
2. Check cross-references between all elements
3. Verify permissions.json completeness
4. Identify opportunities to use latest features from CHANGELOG
5. Check plugin.json validity
")

Phase 5: Present Results to User

CRITICAL: Before asking the user which optimizations to apply, you MUST first output the full findings as readable text. The user needs to understand what was found before making a selection.

Output the consolidated findings as Markdown:

## Identified Optimizations

### Skills (X found)
1. **[Element-Name]**: [Concrete description of what should be optimized and why]
2. ...

### Commands (X found)
1. **[Element-Name]**: [Concrete description]
2. ...

### Agents (X found)
1. **[Element-Name]**: [Concrete description]
2. ...

### Hooks (X found)
1. **[Element-Name]**: [Concrete description]
2. ...

### MCP (X found)
1. **[Element-Name]**: [Concrete description]
2. ...

### Cross-References & Structure (X found)
1. **[Element-Name]**: [Concrete description]
2. ...

Rules:

• Show concrete details for each finding (what, where, why)
• Group by category with count in heading
• Skip categories with zero findings
• Number each finding for easy reference

Phase 6: User Confirmation

After showing the findings, use AskUserQuestion with multiSelect:

• Present all optimizations as checkboxes
• Default: all selected
• Group by agent/category
• Each option text should match the numbered finding above

Phase 7: Execute Approved Optimizations

For each approved optimization:

1. Use the appropriate specialized agent to make the change
2. Track progress with TodoWrite
3. Execute independent changes in parallel

Phase 8: Final Verification

Spawn optimizer-marketplace again for final check:

• All cross-references valid
• No broken links between elements
• Content standards met (no history references)

Specialized Agents

Total docs per agent: 19-43 KB each (vs. 321 KB if loading all)

Content Standards

All agents must enforce:

1. No History References

- Never use "new", "updated", "since vX.Y" - Write timelessly

1. Proper Frontmatter

- All required fields present - Correct types and values

1. Cross-References

- All "Related Skills/Commands" exist - No broken links

Output Format

After completion:

## Marketplace Optimization Complete

### Agent Results
| Agent | Issues Found | Fixed |
|-------|--------------|-------|
| optimizer-skills | X | Y |
| optimizer-commands | X | Y |
| optimizer-agents | X | Y |
| optimizer-hooks | X | Y |
| optimizer-mcp | X | Y |
| optimizer-marketplace | X | Y |

### Changes Made

#### Skills
1. [Change description]

#### Commands
1. [Change description]

#### Agents
1. [Change description]

#### Hooks
1. [Change description]

#### MCP
1. [Change description]

#### Cross-References & Structure
1. [Change description]

### Verification
- [ ] All frontmatter valid
- [ ] No history references
- [ ] All cross-references exist
- [ ] Latest features considered

Related Agents

• optimizer-skills - Skills expert
• optimizer-commands - Commands expert
• optimizer-agents - Agents expert
• optimizer-hooks - Hooks expert
• optimizer-mcp - MCP expert
• optimizer-marketplace - Cross-references & features expert