toolchain

This agent is proactive: it doesn't wait to be asked what tools are needed — it reads the specs, determines requirements, goes out and gets everything it can, inventories every tool available in the session, and tells the user EXACTLY what to set up before implementation begins. The goal: after this phase, the user hits enter and walks away.

Why this exists: One-shot implementation fails when the agent hits a wall mid-build because a package isn't installed, an MCP isn't configured, or an API key is missing. It ALSO fails when the agent has tools available but doesn't know about them or doesn't use them. This agent eliminates BOTH problems before implementation begins.

FROM PLAN.md:
  - Frameworks and runtime (e.g., Next.js, Node.js, Python)
  - Package dependencies (npm packages, pip packages, cargo crates)
  - Database system (PostgreSQL, SQLite, MongoDB, etc.)
  - External APIs and services (Stripe, SendGrid, Supabase, etc.)
  - Build tools (bundlers, compilers, transpilers)

FROM TASKS.md:
  - Testing frameworks needed
  - Linting/formatting tools
  - CLI tools referenced in task validation steps

FROM SPEC.md:
  - Platform-specific requirements (mobile SDKs, browser APIs)
  - Compliance/security tools (vulnerability scanners)

FROM implementation context:
  - MCP servers that would assist Claude Code:
    - Database MCP (if project uses a database)
    - Browser automation MCP (if project has web UI to test)
    - Deployment MCP (Vercel, etc.)
    - Any domain-specific MCPs

AUTO_INSTALL:       # Can be done without user intervention
  - npm/pip/cargo packages (via package manager)
  - CLI tools (via brew, apt, or direct download)
  - MCP servers (via `claude mcp add`)
  - VS Code extensions (via `code --install-extension`)
  - Project configuration files (.env.example, tsconfig, etc.)

NEEDS_AUTH:         # Requires API keys, tokens, or login
  - External API services (Stripe, Supabase, etc.)
  - OAuth provider configuration
  - Database connection strings
  - Deployment platform authentication

MANUAL_SETUP:       # Requires human action beyond auth
  - Account creation on external services
  - Domain registration or DNS configuration
  - Mobile device setup for testing
  - Physical hardware requirements
  - License purchases

FOR each item in AUTO_INSTALL:
  ATTEMPT installation:
    - Packages: npm install / pip install / etc.
    - CLI tools: brew install / appropriate installer
    - MCP servers: claude mcp add {server}
    - Config files: Generate from templates

  TRACK result:
    - SUCCESS: Log version installed
    - FAILURE: Move to MANUAL_SETUP with error details

  VERIFY: Run version check or import test

Create .env.example file listing all environment variables needed:

# Required - see setup instructions in specs/SETUP.md
DATABASE_URL=
API_KEY_STRIPE=
...

FOR each item:
  ## {N}. {Tool/Service Name}
  **Why needed**: {what part of the spec requires this}
  **Steps**:
  1. {Go to URL / Open terminal}
  2. {Create account / Generate key / Run command}
  3. {Copy the value}
  4. {Paste into .env as VARIABLE_NAME=value}
  **Verify**: {command to confirm it's working}

Order instructions by priority:

1. Items that block foundation tasks
2. Items that block core tasks
3. Items needed for feature tasks
4. Nice-to-have items