MCP App StoreMCP App Store
de reprompter
Transforme des prompts désordonnés en prompts structurés et évaluables ou en plans multi-agents ; permet également de rétro-concevoir des résultats exceptionnels en prompts réutilisables.
RePrompter transforme les prompts utilisateurs bruts ou désordonnés en prompts structurés et testables. Il supporte trois modes : Single (nettoyer et noter un seul prompt), Repromptverse (planifier et générer des prompts par agent pour l'orchestration multi-agents), et Reverse (extraire un prompt à partir d'un résultat exemplaire). Inclut des modèles, des flux d'entretien, la génération de critères de succès et des outils d'exécution et d'évaluation.
Utilisez-le lorsque vous devez améliorer la qualité d'un prompt, créer des équipes multi-agents avec des contrats d'artefacts clairs, générer des critères de succès vérifiables ou rétro-concevoir des résultats de haute qualité en prompts reproductibles. À éviter pour les simples questions-réponses ou les tâches d'exécution immédiate.
Conçu pour les runtimes d'agents avancés et les flux de travail de développeurs : Claude Code (tmux/TeamCreate), OpenClaw (sessions_spawn), Codex CLI et agents assistants généraux pour l'ingénierie de prompt et l'orchestration.
RePrompter is a sophisticated prompt-engineering skill with 42 scripts covering four output lanes (Single, Codex Goal, Repromptverse, Reverse). It includes a flywheel learning system, telemetry, strategy learning, and agent orchestration. The SKILL.md is extremely comprehensive at 76K chars but could benefit from progressive disclosure. The shell: true spawnSync in evaluate-outcome.js for judgeCmd is a notable security concern. No scripts ran successfully — JS scripts were skipped by the runner, and shell scripts failed due to missing project context files.
node/npm (required for 39 JS scripts)The spawnSync(judgeCmd, {shell: true}) pattern in evaluate-outcome.js is the main security finding. The comment says 'user-supplied CLI input, not attacker-controlled', which mitigates the risk somewhat, but shell:true with arbitrary input remains a code injection vector if the skill is ever used in a context where judgeCmd could be influenced by external data. The rest of the codebase is well-structured with no exfiltration, no hardcoded secrets, and local-only telemetry. The skill is very ambitious and feature-rich — almost a framework rather than a single skill.
Bounty Hunter — Atlas
Persona skill: 'Atlas' — a profit-focused developer persona for discovering, evaluating and executing paid bounties or freelance tasks with ROI-aware workflows.
Starlark Dev
Create, debug, and test Kurtosis Starlark packages — write package structure, run dry-runs, and inspect plan execution for reliable orchestration.
Maestro Orchestrate
Orchestrates complex engineering workflows with enforced design and approval gates, child-agent delegation, and session state management.
Run Execute
Orchestrates execution of work items across modes (autopilot, confirm, validate) with scripted init/complete tooling, plan/test/report artifacts and strict gati
FastAPI Project Templates
Creates production-ready FastAPI project scaffolds with async patterns, DI, middleware, and testing best practices for high-performance APIs.
Overnight — Autonomous Long-Running Coding
Orchestrates long-running coding goals: decomposes objectives into atomic tasks, dispatches isolated worktree workers, verifies acceptance criteria, and merges
Seedance Prompt — Short
Build, validate, and compress short-form Seedance 2.0 prompts (30–100 words) using a layered compression engine and @Tag delegation for quad-modal (T2V/I2V/V2V/
tmux Remote Control
Control tmux sessions for interactive TTYs: send keystrokes, capture panes, orchestrate parallel coding agents, and monitor output safely.