omc-plan

The plan skill provides a robust framework for transforming broad, ambiguous objectives into precise, actionable work plans. It is designed for developers and project managers who need to avoid scope creep, rework, and missed edge cases by establishing a high-quality foundation before any code is written. By intelligently distinguishing between direct requests and broad inquiries, the agent automatically pivots between direct plan generation and interactive, Socratic-style interview workflows. It utilizes RALPLAN-DR structured deliberation, allowing for iterative loops between a Planner, Architect, and Critic to ensure all requirements are vetted from multiple technical perspectives.

• Intelligent Mode Detection: Automatically switches between Interview mode for scoping broad requests and Direct mode for well-defined tasks.

• Consensus Mode (RALPLAN-DR): Orchestrates a rigorous multi-agent feedback loop (Planner -> Architect -> Critic) to validate plans, supporting both short-form and deliberate execution paths for high-risk changes.

• Requirement Verification: Ensures 80% of claims cite specific file/line references and 90% of acceptance criteria are testable, enforcing professional quality gates.

• Fact-Driven Interviews: Integrates with the codebase via exploration agents to gather context before prompting the user, ensuring questions are informed and efficient.

• Flexible Review Capabilities: Includes standalone review mode to evaluate existing plans, identifying risks and missing edge cases before implementation begins.

• Extensible Provider Support: Allows for custom architecture and critic reviews using external CLI tools like codex if needed, ensuring modular, expert-level analysis.

• Use this skill when initiating complex projects, tackling vague high-level requirements, or preparing for high-stakes modifications like production migrations, security changes, or API refactors.

• The interview process is designed for sequential, one-at-a-time questioning to prevent user cognitive overload; each response is used to refine the scope of the next inquiry.

• State management is handled through a persistent lifecycle, ensuring that multi-step consensus loops are tracked effectively via session IDs without interfering with subsequent execution modes.

• For high-risk operations involving security, compliance, or irreversible data changes, utilize the --deliberate flag to trigger deep-dive analysis modes.

• Integration with the broader multi-agent environment allows for seamless handoffs to execution agents like ralph or executor once the final plan is approved and ready for development.