**OVERVIEW:**
This epic implements a comprehensive role enforcement system to ensure agents strictly adhere to their designated responsibilities when using the "run-claude" command.
**CURRENT PROBLEM:**
- Agents can perform tasks outside their designated roles
- No system prompts constrain agent behavior based on role
- All agents have access to all MCP tools regardless of role
- No validation prevents role boundary violations
**ROLE DEFINITIONS:**
**Orchestrator:**
- Strategic planning and task decomposition
- Coordination between agents
- Monitoring and unblocking workers
- NEVER implements code directly
- Tools: Planning, coordination, memory search/store
**Worker:**
- Code implementation and testing
- Bug fixes and feature development
- NEVER does strategic planning or coordination
- Tools: Code editing, testing, file operations
**Reviewer:**
- Code review and quality assurance
- Testing and validation
- NEVER implements new features
- Tools: Review, audit, test execution
**SUCCESS CRITERIA:**
1. Each agent type has role-specific system prompts
2. MCP tool access is restricted by role
3. Validation prevents role violations
4. Clear error messages guide agents back to their role
5. Comprehensive test suite verifies compliance
**TECHNICAL APPROACH:**
1. Modify run-claude command to inject role-specific prompts
2. Implement MCP tool access control layer
3. Add role validation middleware
4. Create guardrails with helpful error messages
5. Build test suite for role compliance
**OVERVIEW:**
This epic implements a comprehensive role enforcement system to ensure agents strictly adhere to their designated responsibilities when using the "run-claude" command.
**CURRENT PROBLEM:**
- Agents can perform tasks outside their designated roles
- No system prompts constrain agent behavior based on role
- All agents have access to all MCP tools regardless of role
- No validation prevents role boundary violations
**ROLE DEFINITIONS:**
**Orchestrator:**
- Strategic planning and task decomposition
- Coordination between agents
- Monitoring and unblocking workers
- NEVER implements code directly
- Tools: Planning, coordination, memory search/store
**Worker:**
- Code implementation and testing
- Bug fixes and feature development
- NEVER does strategic planning or coordination
- Tools: Code editing, testing, file operations
**Reviewer:**
- Code review and quality assurance
- Testing and validation
- NEVER implements new features
- Tools: Review, audit, test execution
**SUCCESS CRITERIA:**
1. Each agent type has role-specific system prompts
2. MCP tool access is restricted by role
3. Validation prevents role violations
4. Clear error messages guide agents back to their role
5. Comprehensive test suite verifies compliance
**TECHNICAL APPROACH:**
1. Modify run-claude command to inject role-specific prompts
2. Implement MCP tool access control layer
3. Add role validation middleware
4. Create guardrails with helpful error messages
5. Build test suite for role compliance