---

name: foundry description: Three-stage code implementation with planning. Gemini (Engineer) plans, Claude (Caster) implements, Gemini (Inspector) verifies. ~3x token cost for complex, multi-file changes.

The Foundry: Code Implementation Triple

When to Activate

Use this skill when the user requests:

• Complex multi-file implementations
• Architectural changes requiring planning
• Features with significant design decisions
• Tasks where upfront planning reduces rework
• User explicitly mentions "foundry", "plan and implement", or "complex implementation"

Architecture

Triple: Engineer + Caster + Inspector

The Engineer (Gemini 3 Pro)

• Role: Planner/Architect
• Analyzes requirements and codebase context
• Creates implementation blueprint with steps
• Identifies files to modify and dependencies
• Sets success criteria for verification

The Caster (Claude)

• Role: Implementer
• Executes the Engineer's blueprint
• Writes production-ready code
• Follows existing patterns in codebase

The Inspector (Gemini 3 Pro)

• Role: Verifier
• Compares implementation against blueprint
• Verifies all requirements met
• Checks for bugs, security, edge cases
• Returns PASS or identifies gaps

Execution Flow

Copy[User Request]
    ↓
[Engineer] → Creates blueprint
    ↓
[Caster] → Implements code
    ↓
[Inspector] → Verifies against blueprint
    ↓
    ├── PASS → Return code to user
    └── FAIL → Caster fixes → Inspector re-verifies

Invocation

Copypython3 ~/.claude/skills/foundry/foundry.py "Your complex implementation task"

Options

Copypython3 ~/.claude/skills/foundry/foundry.py "Your task" --max-iterations 3 --verbose --skip-planning

Example Usage

User: "Add user authentication with JWT tokens"

Copypython3 ~/.claude/skills/foundry/foundry.py "Add JWT authentication to the API - create login endpoint, middleware for protected routes, and token refresh logic"

Why This Works

1. Upfront Planning: Engineer catches design issues before coding
2. Cross-Model Verification: Gemini plans and verifies, Claude implements
3. Blueprint Alignment: Inspector checks against original plan, not just code quality
4. Fewer Iterations: Good planning reduces fix cycles

Token Efficiency

• ~3x base cost (1 plan + 1 implementation + 1 verification)
• Best for: Multi-file changes, new features, architectural work
• Avoid for: Simple bug fixes (use Forge instead)

When to Use Forge vs Foundry

Requirements

• Claude Code CLI installed and authenticated (claude command available)
• Gemini CLI installed (gemini command available)
• Python package: rich (for formatted output)