Dev prompt for Multi-Agent System Design for Admin Dashboard

As a Principal Architect, design a multi-agent system, leveraging the Model Context Protocol (MCP), to fully automate the implementation of a new feature within a full-stack TypeScript monorepo.

Project Context:
- Monorepo Stack: Next.js 15 (frontend), NestJS (backend).
- Database: PostgreSQL, using Drizzle ORM for schema definitions and migrations.
- UI Framework: Tailwind CSS v4.
- Access Control: Role-Based Access Control (RBAC) implemented with CASL.
- Development Standards: Agents must strictly adhere to project-specific STYLE_GUIDE.md and AGENTS.md for all coding and interaction guidelines.

Feature Request:
Automate the complete development of an administrative dashboard for managing user roles and permissions.

Agent Architecture & Responsibilities:
1. Planner Agent:
- Input: The user story (feature request).
- Task: Deconstruct the user story into granular, actionable sub-tasks.
- Output: A tasks.json file detailing all required routes, database schema changes, UI components, and associated tests.
2. Coding Agent:
- Input: The tasks.json plan from the Planner Agent, and error logs from the Reviewer Agent.
- Task: Generate all necessary code for the specified feature, strictly following STYLE_GUIDE.md and AGENTS.md. console.log is permitted for debugging purposes only; no production logging.
- Constraint: If the agent enters a development loop, failing to progress after 5 attempts, it must generate a DEBUG_LOG.md explaining the failure mode and cease operations.
3. Reviewer Agent:
- Input: Code artifacts produced by the Coding Agent.
- Task: Execute npm run lint and npm run test on the generated codebase.
- Output: If tests or linting fail, forward the error log back to the Coding Agent for remediation. If successful, pass the validated code to the QA Agent.
4. QA Agent:
- Input: Verified code from the Reviewer Agent.
- Task: Develop comprehensive Playwright E2E tests for the new UI components and ensure their successful execution in headless mode.
- Output: A final pass/fail status for the E2E tests.

Overall Constraint:
- The entire development process must require zero manual coding intervention from the user.

Deliverables:
Provide the following:
1. System Prompt for Each Agent: A distinct, detailed system prompt for each of the four agent types (Planner, Coding, Reviewer, QA).
2. MCP Server Configuration: The necessary Model Context Protocol (MCP) server configuration to grant agents secure access to the file system, terminal, and PostgreSQL database.