MCP App StoreMCP App Store
de tmdgusya
Modèle itératif de Q&R et d'exploration de code pour transformer des demandes d'utilisateurs vagues en un Brief de Contexte clair.
Affine les demandes vagues des utilisateurs en un travail bien défini en exécutant une boucle itérative de Q&R (une question par message) tout en déployant des sous-agents d'exploration de code en parallèle. Produit un Brief de Contexte contenant l'Objectif, le Périmètre, le Contexte Technique, les Contraintes, les Critères de Succès, les Questions Ouvertes et une Évaluation de la Complexité pour orienter les prochaines étapes.
Règles fondamentales :
Exemples de déclencheurs :
Repository : https://github.com/tmdgusya/roach-pi
Agentic Clarification is a pure-instruction skill (no scripts) that guides an agent through iterative Q&A and optional codebase exploration to resolve vague user requests into a structured Goal Contract. Well-documented with clear triggers, anti-patterns, red flags, and a minimal checklist. Assumes a specific runtime ecosystem (Goal Contract, clarification_state tool) that may not be universally available. No security concerns — no code execution, no network calls, no credentials.
Clean instructional skill with no executable code. Tightly coupled to the roach-pi agentic ecosystem (Goal Contract runtime, clarification_state tool, explorer subagent). Useful within that ecosystem but less portable as a standalone skill.
Maestro Orchestrate
Orchestrates complex engineering workflows with enforced design and approval gates, child-agent delegation, and session state management.
Execution Plan Splitter
Break large goals into 30/60/90-day execution plans with milestones, resource needs, and stop/go criteria for practical roadmaps.
Writing Implementation Plans
Generate detailed, engineer-ready implementation plans from design docs: file paths, step-by-step tasks, tests, and verification instructions.
Build Unplanned Feature
Orchestrates ad-hoc feature development: requirement gathering, planning, implementation and validation across task-planning, implementation, and validation loo
Plan Design Review
Interactive, designer-eye review of design plans: rates dimensions, finds gaps, and edits the plan to add missing UI decisions and interaction states.
Ideate — Discovery Brief
Create a concise Discovery Brief that frames a task, lists constraints, states testable hypotheses, and surfaces open questions for downstream research and plan
Plan Mode
Produce a concrete, actionable markdown plan saved to .clawg/plans/ — planning-only mode that avoids execution or file mutations outside the plan file.
Writing Plans
Create precise, bite-sized implementation plans from a spec so engineers can implement features reliably using TDD and exact file/command instructions.