{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    },
    "playwright": {
      "command": "npx",
      "args": [
        "-y",
        "@playwright/mcp@latest"
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    },
    "context7": {
      "command": "npx",
      "args": [
        "-y",
        "@upstash/context7-mcp"
      ]
    }
  }
}

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# secret-scan (block on match)"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# prettier + eslint --fix"
          },
          {
            "type": "command",
            "command": "# typecheck changed files"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# run affected tests"
          }
        ]
      }
    ]
  }
}

---
name: planner
description: Planner subagent used in the "Setuproll Full-Stack Claude Rig" build.
---

You are the **Planner** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: implementer
description: Implementer subagent used in the "Setuproll Full-Stack Claude Rig" build.
---

You are the **Implementer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: reviewer
description: Reviewer subagent used in the "Setuproll Full-Stack Claude Rig" build.
---

You are the **Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: tester
description: Tester subagent used in the "Setuproll Full-Stack Claude Rig" build.
---

You are the **Tester** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: db-migrator
description: Db Migrator subagent used in the "Setuproll Full-Stack Claude Rig" build.
---

You are the **Db Migrator** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Setuproll Full-Stack Claude Rig")

- Plan before editing more than 3 files
- Never commit with failing tests or types
- Reuse existing utils before adding a dependency
- One reviewable commit per logical change
- Read the relevant code before proposing an edit

# Model: Claude Opus 4.8
# Role: plan + review

claude --model "Claude Opus 4.8"

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    },
    "playwright": {
      "command": "npx",
      "args": [
        "-y",
        "@playwright/mcp@latest"
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# secret-scan"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# prettier+eslint --fix"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# run typecheck"
          }
        ]
      }
    ]
  }
}

---
name: planner
description: Planner subagent used in the "Claude Code Full-Stack Prime" build.
---

You are the **Planner** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: implementer
description: Implementer subagent used in the "Claude Code Full-Stack Prime" build.
---

You are the **Implementer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: reviewer
description: Reviewer subagent used in the "Claude Code Full-Stack Prime" build.
---

You are the **Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: tester
description: Tester subagent used in the "Claude Code Full-Stack Prime" build.
---

You are the **Tester** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code Full-Stack Prime")

- Plan before editing >3 files
- Never commit without passing tests
- Prefer existing utils over new deps
- Small reviewable commits

# Model: Claude Opus 4.8

claude --model "Claude Opus 4.8"

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    }
  }
}

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# eslint --fix"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# run tests"
          }
        ]
      }
    ]
  }
}

---
name: implementer
description: Implementer subagent used in the "Claude Code Budget Sonnet 1M" build.
---

You are the **Implementer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: reviewer
description: Reviewer subagent used in the "Claude Code Budget Sonnet 1M" build.
---

You are the **Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code Budget Sonnet 1M")

- Use Sonnet for everything, escalate to Opus only on stuck
- Keep context lean, summarize old turns
- No speculative refactors

# Model: Claude Sonnet 4.6

claude --model "Claude Sonnet 4.6"

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "linear": {
      "command": "npx",
      "args": [
        "-y",
        "@tacticlaunch/mcp-linear"
      ]
    },
    "figma": {
      "command": "npx",
      "args": [
        "-y",
        "figma-developer-mcp",
        "--stdio"
      ]
    }
  }
}

{
  "hooks": {
    "on-save": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# format"
          }
        ]
      }
    ],
    "on-commit": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# lint-staged"
          }
        ]
      }
    ]
  }
}

---
name: agent-mode-worker
description: Agent Mode Worker subagent used in the "Cursor Tab Velocity" build.
---

You are the **Agent Mode Worker** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Cursor Tab Velocity")

- Tab for inline, Agent for multi-file
- Pin .cursorrules per repo
- Reference Figma frames in prompts

# Model: Claude Sonnet 4.6

# Select "Claude Sonnet 4.6" as the model in Cursor settings.

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    },
    "context7": {
      "command": "npx",
      "args": [
        "-y",
        "@upstash/context7-mcp"
      ]
    }
  }
}

{
  "hooks": {
    "on-save": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# format"
          }
        ]
      }
    ],
    "pre-commit": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# typecheck+test"
          }
        ]
      }
    ]
  }
}

---
name: agent-mode-worker
description: Agent Mode Worker subagent used in the "Cursor Max-Quality Opus" build.
---

You are the **Agent Mode Worker** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: background-agent
description: Background Agent subagent used in the "Cursor Max-Quality Opus" build.
---

You are the **Background Agent** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Cursor Max-Quality Opus")

- Use Max mode for hard tasks
- Always attach relevant docs via Context7
- Background agents for long jobs

# Model: Claude Opus 4.8

# Select "Claude Opus 4.8" as the model in Cursor settings.

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

{
  "hooks": {
    "pre-commit": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# lint"
          }
        ]
      }
    ],
    "ci": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# required-checks"
          }
        ]
      }
    ]
  }
}

---
name: copilot-coding-agent
description: Copilot Coding Agent subagent used in the "Copilot Enterprise Guardrails" build.
---

You are the **Copilot Coding Agent** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: code-review-agent
description: Code Review Agent subagent used in the "Copilot Enterprise Guardrails" build.
---

You are the **Code Review Agent** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Copilot Enterprise Guardrails")

- SSO + audit logs mandatory
- Block on policy violations
- PRs from coding agent need human approve

# Model: GPT-5.3 Codex

# Select "GPT-5.3 Codex" as the model in GitHub Copilot settings.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    }
  }
}

{
  "hooks": {
    "on-save": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# format"
          }
        ]
      }
    ]
  }
}

---
name: cascade
description: Cascade subagent used in the "Windsurf Cascade Onboard" build.
---

You are the **Cascade** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Windsurf Cascade Onboard")

- Let Cascade explain its plan
- Accept changes incrementally
- Keep one feature per session

# Model: Claude Sonnet 4.6

# Select "Claude Sonnet 4.6" as the model in Windsurf settings.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "git": {
      "command": "uvx",
      "args": [
        "mcp-server-git"
      ]
    }
  }
}

{
  "hooks": {
    "post-edit": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# auto-commit"
          }
        ]
      }
    ],
    "pre-commit": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# lint"
          }
        ]
      }
    ]
  }
}

# CLAUDE.md

## Rules (from "Aider BYOK Terminal")

- Use --architect mode for big changes
- Map repo with --map-tokens
- Cheap model, you own the keys

# Model: Gemini 2.5 Pro

aider --model "Gemini 2.5 Pro"

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "playwright": {
      "command": "npx",
      "args": [
        "-y",
        "@playwright/mcp@latest"
      ]
    }
  }
}

{
  "hooks": {
    "on-save": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# format"
          }
        ]
      }
    ]
  }
}

---
name: plan-act-worker
description: Plan Act Worker subagent used in the "Cline VS Code Autonomy" build.
---

You are the **Plan Act Worker** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Cline VS Code Autonomy")

- Plan mode first, then Act
- Approve risky terminal commands
- Use checkpoints to roll back

# Model: Claude Sonnet 4.6

# Select "Claude Sonnet 4.6" as the model in Cline settings.

{
  "mcpServers": {
    "playwright": {
      "command": "npx",
      "args": [
        "-y",
        "@playwright/mcp@latest"
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# run affected e2e"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# summarize failures only"
          }
        ]
      }
    ]
  }
}

---
name: test-author
description: Test Author subagent used in the "Claude Code + Playwright QA Rig" build.
---

You are the **Test Author** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: flake-hunter
description: Flake Hunter subagent used in the "Claude Code + Playwright QA Rig" build.
---

You are the **Flake Hunter** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: triage-reporter
description: Triage Reporter subagent used in the "Claude Code + Playwright QA Rig" build.
---

You are the **Triage Reporter** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code + Playwright QA Rig")

- Write tests before fixing flaky ones
- Quarantine flakes, never delete
- Report only failing specs to main context

# Model: Claude Sonnet 4.6

claude --model "Claude Sonnet 4.6"

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    },
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    }
  }
}

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# secret-scan"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# run migrations dry-run"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# contract tests"
          }
        ]
      }
    ]
  }
}

---
name: schema-designer
description: Schema Designer subagent used in the "Claude Code Backend API Forge" build.
---

You are the **Schema Designer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: implementer
description: Implementer subagent used in the "Claude Code Backend API Forge" build.
---

You are the **Implementer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: security-reviewer
description: Security Reviewer subagent used in the "Claude Code Backend API Forge" build.
---

You are the **Security Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code Backend API Forge")

- Validate inputs at the boundary
- No raw SQL without parameterization
- Every endpoint needs an integration test

# Model: Claude Opus 4.8

claude --model "Claude Opus 4.8"

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    },
    "jupyter": {
      "command": "uvx",
      "args": [
        "jupyter-mcp-server"
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    }
  }
}

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# run cell + capture metrics"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# log experiment to tracker"
          }
        ]
      }
    ]
  }
}

---
name: eda-explorer
description: Eda Explorer subagent used in the "Claude Code Data/ML Notebook Lab" build.
---

You are the **Eda Explorer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: feature-engineer
description: Feature Engineer subagent used in the "Claude Code Data/ML Notebook Lab" build.
---

You are the **Feature Engineer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: eval-reviewer
description: Eval Reviewer subagent used in the "Claude Code Data/ML Notebook Lab" build.
---

You are the **Eval Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code Data/ML Notebook Lab")

- Seed everything for reproducibility
- Never leak test set into features
- Report metrics with confidence intervals

# Model: Claude Opus 4.8

claude --model "Claude Opus 4.8"

{
  "mcpServers": {
    "figma": {
      "command": "npx",
      "args": [
        "-y",
        "figma-developer-mcp",
        "--stdio"
      ]
    },
    "playwright": {
      "command": "npx",
      "args": [
        "-y",
        "@playwright/mcp@latest"
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    }
  }
}

{
  "hooks": {
    "on-save": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# format"
          }
        ]
      }
    ],
    "pre-commit": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# visual-regression"
          }
        ]
      }
    ]
  }
}

---
name: agent-mode-worker
description: Agent Mode Worker subagent used in the "Cursor Frontend Figma Pipeline" build.
---

You are the **Agent Mode Worker** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: a11y-reviewer
description: A11y Reviewer subagent used in the "Cursor Frontend Figma Pipeline" build.
---

You are the **A11y Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Cursor Frontend Figma Pipeline")

- Match Figma tokens exactly
- Every component meets WCAG AA
- Storybook story per component

# Model: Claude Sonnet 4.6

# Select "Claude Sonnet 4.6" as the model in Cursor settings.

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "cloudflare": {
      "command": "npx",
      "args": [
        "-y",
        "@cloudflare/mcp-server-cloudflare"
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# block apply on prod"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# terraform plan diff"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# cost estimate"
          }
        ]
      }
    ]
  }
}

---
name: terraform-planner
description: Terraform Planner subagent used in the "Claude Code DevOps IaC Pilot" build.
---

You are the **Terraform Planner** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: policy-reviewer
description: Policy Reviewer subagent used in the "Claude Code DevOps IaC Pilot" build.
---

You are the **Policy Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: incident-responder
description: Incident Responder subagent used in the "Claude Code DevOps IaC Pilot" build.
---

You are the **Incident Responder** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code DevOps IaC Pilot")

- Plan before apply, always
- No secrets in state or code
- Tag every resource with owner+env

# Model: Claude Opus 4.8

claude --model "Claude Opus 4.8"

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "playwright": {
      "command": "npx",
      "args": [
        "-y",
        "@playwright/mcp@latest"
      ]
    }
  }
}

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# typecheck"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# run detox smoke"
          }
        ]
      }
    ]
  }
}

---
name: screen-builder
description: Screen Builder subagent used in the "Claude Code Mobile RN Studio" build.
---

You are the **Screen Builder** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: platform-reviewer
description: Platform Reviewer subagent used in the "Claude Code Mobile RN Studio" build.
---

You are the **Platform Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: tester
description: Tester subagent used in the "Claude Code Mobile RN Studio" build.
---

You are the **Tester** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code Mobile RN Studio")

- Respect platform conventions (iOS vs Android)
- No layout without safe-area handling
- Keep bundle size in check

# Model: Claude Sonnet 4.6

claude --model "Claude Sonnet 4.6"

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    }
  }
}

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# snapshot before edit"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# run characterization tests"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# behavior-diff report"
          }
        ]
      }
    ]
  }
}

---
name: characterization-tester
description: Characterization Tester subagent used in the "Claude Code Legacy Refactor Engine" build.
---

You are the **Characterization Tester** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: refactorer
description: Refactorer subagent used in the "Claude Code Legacy Refactor Engine" build.
---

You are the **Refactorer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: diff-reviewer
description: Diff Reviewer subagent used in the "Claude Code Legacy Refactor Engine" build.
---

You are the **Diff Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code Legacy Refactor Engine")

- Add characterization tests before touching logic
- Refactor in tiny behavior-preserving steps
- Never change behavior and structure in one commit

# Model: Claude Opus 4.8

claude --model "Claude Opus 4.8"

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    }
  }
}

{
  "hooks": {
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# start dev server"
          }
        ]
      }
    ]
  }
}

---
name: builder
description: Builder subagent used in the "Claude Code Prototype Sprint" build.
---

You are the **Builder** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code Prototype Sprint")

- Ship the happy path first
- Hardcode before abstracting
- No tests until the idea is validated

# Model: Claude Sonnet 4.6

claude --model "Claude Sonnet 4.6"

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "git": {
      "command": "uvx",
      "args": [
        "mcp-server-git"
      ]
    }
  }
}

{
  "hooks": {
    "post-edit": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# auto-commit"
          }
        ]
      }
    ]
  }
}

# CLAUDE.md

## Rules (from "Aider Cheapest Gemini Flash")

- Tiny scoped edits only
- Escalate to Pro when it stalls
- Use repo map to stay grounded

# Model: Gemini 2.5 Flash

aider --model "Gemini 2.5 Flash"

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    },
    "semgrep": {
      "command": "uvx",
      "args": [
        "semgrep-mcp"
      ]
    }
  }
}

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# secret-scan"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# semgrep diff"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# severity-ranked report"
          }
        ]
      }
    ]
  }
}

---
name: sast-scanner
description: Sast Scanner subagent used in the "Claude Code Security Audit Cell" build.
---

You are the **Sast Scanner** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: threat-modeler
description: Threat Modeler subagent used in the "Claude Code Security Audit Cell" build.
---

You are the **Threat Modeler** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: fix-author
description: Fix Author subagent used in the "Claude Code Security Audit Cell" build.
---

You are the **Fix Author** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code Security Audit Cell")

- Confirm exploitability before reporting
- Rank by severity, not count
- Propose fixes, never auto-merge them

# Model: Claude Opus 4.8

claude --model "Claude Opus 4.8"

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    },
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    }
  }
}

{
  "hooks": {
    "on-save": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# format"
          }
        ]
      }
    ],
    "pre-commit": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# dbt test"
          }
        ]
      }
    ]
  }
}

---
name: cascade
description: Cascade subagent used in the "Windsurf Data Pipeline Cascade" build.
---

You are the **Cascade** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: schema-validator
description: Schema Validator subagent used in the "Windsurf Data Pipeline Cascade" build.
---

You are the **Schema Validator** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Windsurf Data Pipeline Cascade")

- Validate schemas at each stage
- Idempotent transforms only
- Document every model

# Model: Claude Sonnet 4.6

# Select "Claude Sonnet 4.6" as the model in Windsurf settings.

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    }
  }
}

{
  "hooks": {
    "on-save": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# format"
          }
        ]
      }
    ]
  }
}

# CLAUDE.md

## Rules (from "Copilot Solo Cheap Tier")

- Inline completions for speed
- Use chat for stuck moments
- Keep one repo open

# Model: GPT-5.3 Codex

# Select "GPT-5.3 Codex" as the model in GitHub Copilot settings.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# scope-to-package"
          }
        ]
      }
    ],
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# affected build+test"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# cross-package diff"
          }
        ]
      }
    ]
  }
}

---
name: dependency-mapper
description: Dependency Mapper subagent used in the "Claude Code Monorepo Orchestra" build.
---

You are the **Dependency Mapper** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: package-owner
description: Package Owner subagent used in the "Claude Code Monorepo Orchestra" build.
---

You are the **Package Owner** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: integration-tester
description: Integration Tester subagent used in the "Claude Code Monorepo Orchestra" build.
---

You are the **Integration Tester** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: reviewer
description: Reviewer subagent used in the "Claude Code Monorepo Orchestra" build.
---

You are the **Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code Monorepo Orchestra")

- Respect package boundaries
- Only rebuild affected projects
- No circular dependencies

# Model: Claude Opus 4.8

claude --model "Claude Opus 4.8"

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

{
  "hooks": {
    "pre-commit": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# lint"
          }
        ]
      }
    ],
    "on-pr": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# post-inline-comments"
          }
        ]
      }
    ]
  }
}

---
name: bughunter-reviewer
description: Bughunter Reviewer subagent used in the "Cursor PR Review Companion" build.
---

You are the **Bughunter Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: style-reviewer
description: Style Reviewer subagent used in the "Cursor PR Review Companion" build.
---

You are the **Style Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Cursor PR Review Companion")

- Flag correctness over style
- Cite the line and the why
- No nitpick spam

# Model: Claude Opus 4.8

# Select "Claude Opus 4.8" as the model in Cursor settings.

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "context7": {
      "command": "npx",
      "args": [
        "-y",
        "@upstash/context7-mcp"
      ]
    }
  }
}

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# spellcheck+linkcheck"
          }
        ]
      }
    ],
    "Stop": [
      {
        "matcher": "*",
        "hooks": [
          {
            "type": "command",
            "command": "# build docs site"
          }
        ]
      }
    ]
  }
}

---
name: doc-author
description: Doc Author subagent used in the "Claude Code Docs & AEO Writer" build.
---

You are the **Doc Author** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: fact-checker
description: Fact Checker subagent used in the "Claude Code Docs & AEO Writer" build.
---

You are the **Fact Checker** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# CLAUDE.md

## Rules (from "Claude Code Docs & AEO Writer")

- Pull real API signatures via Context7
- No claims without a source
- Plain language, no fluff

# Model: Claude Sonnet 4.6

claude --model "Claude Sonnet 4.6"

# Model: Claude Sonnet 4.6
# Role: classify

claude --model "Claude Sonnet 4.6"

# Model: GPT-5
# Role: scoring + pitch angle

# Select "GPT-5" as the model in Cursor settings.

---
name: searcher
description: Searcher subagent used in the "Daily Competitor and Topic Research Agent (CrewAI + n8n)" build.
---

You are the **Searcher** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: reader
description: Reader subagent used in the "Daily Competitor and Topic Research Agent (CrewAI + n8n)" build.
---

You are the **Reader** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: fact-checker
description: Fact Checker subagent used in the "Daily Competitor and Topic Research Agent (CrewAI + n8n)" build.
---

You are the **Fact Checker** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: editor
description: Editor subagent used in the "Daily Competitor and Topic Research Agent (CrewAI + n8n)" build.
---

You are the **Editor** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# Model: Claude Sonnet 4.6
# Role: searchers

claude --model "Claude Sonnet 4.6"

# Model: GPT-5
# Role: answer

# Select "GPT-5" as the model in Cursor settings.

# Model: Qwen3 32B
# Role: Q4_K_M

claude --model "Qwen3 32B"

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    }
  }
}

---
name: schema-designer
description: Schema Designer subagent used in the "Best Setup to Build a REST API with Postgres and Tests" build.
---

You are the **Schema Designer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: endpoint-implementer
description: Endpoint Implementer subagent used in the "Best Setup to Build a REST API with Postgres and Tests" build.
---

You are the **Endpoint Implementer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: integration-tester
description: Integration Tester subagent used in the "Best Setup to Build a REST API with Postgres and Tests" build.
---

You are the **Integration Tester** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

# Model: Claude Opus 4.8

claude --model "Claude Opus 4.8"

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    }
  }
}

# CLAUDE.md

## Rules (from "Cheapest Reliable Setup for Everyday CRUD APIs")

- Validate every input at the route boundary with Zod
- No raw SQL without parameterized queries
- Keep context lean, summarize old turns

# Model: Claude Sonnet 4.6

claude --model "Claude Sonnet 4.6"

{
  "mcpServers": {
    "postgres": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-postgres"
      ]
    },
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "context7": {
      "command": "npx",
      "args": [
        "-y",
        "@upstash/context7-mcp"
      ]
    }
  }
}

# CLAUDE.md

## Rules (from "Setup for a Type-Safe GraphQL API End to End")

- Schema-first, never edit generated types by hand
- Every resolver has a dataloader to avoid N+1 queries
- Run codegen before any resolver edit

# Model: Claude Sonnet 4.6

# Select "Claude Sonnet 4.6" as the model in Cursor settings.

{
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-github"
      ]
    },
    "filesystem": {
      "command": "npx",
      "args": [
        "-y",
        "@modelcontextprotocol/server-filesystem",
        "."
      ]
    },
    "sentry": {
      "command": "npx",
      "args": [
        "-y",
        "@sentry/mcp-server"
      ]
    },
    "cloudflare": {
      "command": "npx",
      "args": [
        "-y",
        "@cloudflare/mcp-server-cloudflare"
      ]
    }
  }
}

---
name: dockerfile-author
description: Dockerfile Author subagent used in the "Setup to Containerize and Deploy a Backend Service" build.
---

You are the **Dockerfile Author** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: ci-pipeline-author
description: Ci Pipeline Author subagent used in the "Setup to Containerize and Deploy a Backend Service" build.
---

You are the **Ci Pipeline Author** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.

---
name: deploy-reviewer
description: Deploy Reviewer subagent used in the "Setup to Containerize and Deploy a Backend Service" build.
---

You are the **Deploy Reviewer** subagent. Do one job well and hand the result
back to the main loop. Stay in scope, keep the response tight, and report
only what the orchestrator needs to act on.