[Skill Name]

General Description

This skill is the main orchestrator (Architect) and the user's sole interaction point. Responsible for the entire document-to-skill automation workflow. Progressive disclosure: the agent first reads the description to determine triggering; if relevant, loads the main instructions; calls sub-agents for complex steps.

Workflow Steps

Step 1: Collect Raw Materials

• Create temporary directory temp-skills/temp-jsons to store the generated all-analyses.json and meta-blueprint.json
• Based on user-provided documentation directory (e.g., /vercel-ai-sdk/docs), if no relevant directory is found, do not use search tools; first confirm the directory file address with the user.
• If total documentation files exceed 40, stop immediately and reply: "I detected {{actual_count}} files—over the 40-file limit, which dilutes context and hurts SKILL.md quality. Please split into smaller batches (≤15 files each);
• Collect all .md original documentation files, use cp to copy to temp-skills/references directory
• For each document in the references directory, call a separate instance of <doc_analyzer_agent> in parallel, but strictly limit each agent to read and analyze only its assigned single file.
• Generate all-analyses.json (summary JSON including summary, toc, key_apis, etc.)
• Use mv to rename the temporary directory temp-skills/ to: [new-skill-name]/ (using the user's final confirmed skill name)

Step 2: Generate Plan (Planning Phase)

• Present all-analyses.json summary to the user
• Based on analysis results, generate multiple skill plan options for user selection:
  • Option A: Complete SDK Skill - Large comprehensive skill containing all modules
  • Option B: Core Functionality Skill - Streamlined skill focusing on most commonly used APIs
  • Option C: Modular Skill Set - Multiple small skills split by functional domain
• For each option provide:
  • Skill description and intended use
  • Expected file structure
  • Main functionality coverage
  • Recommended usage scenarios
• User selects an option or requests hybrid customization

Step 3: Design Blueprint (User Collaboration)

• Based on user-selected plan, collect detailed configuration:
  • Skill name (final confirmation)
  • Skill description (following third-person standard)
  • Module planning ("divide core.md into 'core' module")
  • Routing logic ("queries containing useChat route to 'ui' module")
  • Cross-module patterns ("define a complete 'Core + UI' workflow")
  • Progressive disclosure strategy (what content goes into references/)
• Generate meta-blueprint.json (blueprint JSON)
• Users generally won't directly view JSON files, so generate a summary for user confirmation; if user is unsatisfied, prompt for iteration
• Example meta-blueprint.json:

  {