Ralplan

Funktionen

• Consensus planning workflow
• Pre-execution gate for vague requests
• Automated agent review loop (Planner, Architect, Critic)
• User-interactive or fully automated planning modes
• Deliberate mode for high-risk tasks

Anwendungsfälle

• When a request is too vague for direct execution (e.g., 'improve the app').
• Before undertaking complex tasks requiring architectural review and consensus.
• To ensure clear acceptance criteria and test specifications are defined upfront.
• When requiring explicit user approval at key decision points in the planning process.

Nicht-Ziele

• Directly executing code or making system changes without prior planning and approval.
• Handling simple, well-defined tasks that do not require multi-agent consensus.
• Replacing the need for user input or oversight in critical decision-making points (when `--interactive` is not used).

Workflow

1. User submits a request.
2. Gate detects if the request is too vague for direct execution.
3. If vague, request is redirected to 'ralplan' for consensus planning.
4. Planner creates an initial plan and summary.
5. Architect reviews the plan for soundness.
6. Critic evaluates the plan against quality criteria.
7. Iterative review loop (max 5 iterations) until Critic approves.
8. On approval, plan is marked 'pending approval' (or proceeds to execution if interactive approval is captured).
9. User may choose execution path (team or ralph) if interactive mode is enabled.

Praktiken

• Consensus Planning
• Pre-Execution Gating
• Multi-Agent Collaboration
• Risk Management

Voraussetzungen

• Claude Code environment
• oh-my-claudecode installed
• Minimum Claude subscription or API key