Task 1: System Prompt Implementation for Role Enforcement

Tinker

Home Epics Task 1: System Prompt Implementation for Role EnforcementEdit

Cancel

Title *

Project *

Description

**OBJECTIVE:**
Implement role-specific system prompts that are injected when using the "run-claude" command to constrain each agent type to their specific duties.

**REQUIREMENTS:**

1. **Orchestrator System Prompt:**
   - Emphasize coordination and planning only
   - Explicitly prohibit direct code implementation
   - Focus on task decomposition, delegation, and monitoring
   - Include clear boundaries: "You MUST NOT write code directly"

2. **Worker System Prompt:**
   - Focus on implementation and testing
   - Prohibit strategic planning or task decomposition
   - Emphasize following orchestrator-defined tasks
   - Include boundary: "You MUST NOT create new tasks or reorganize work"

3. **Reviewer System Prompt:**
   - Focus on code review and quality assurance
   - Prohibit implementing new features
   - Emphasize audit, testing, and validation
   - Include boundary: "You MUST NOT implement solutions, only review them"

**TECHNICAL DETAILS:**
- Modify the run-claude command to detect agent type
- Create template system prompts for each role
- Inject prompts before agent interaction begins
- Ensure prompts are persistent throughout the session

**FILES TO MODIFY:**
- `/app/commands/run_claude.rb` (or equivalent)
- Create `/lib/tinker/role_prompts.rb` for prompt templates

**ACCEPTANCE CRITERIA:**
- Each agent type receives correct system prompt
- Prompts are clearly displayed at session start
- Agents reference role constraints in their responses
- No agent can perform tasks outside their designated role