Vibe Check

소개

Vibe Check functions as an AI meta-mentor for autonomous agents, implementing Anthropic's Model Context Protocol (MCP). It utilizes Chain-Pattern Interrupts (CPI) to intervene at critical moments, preventing pattern inertia, over-engineering, and Reasoning Lock-In (RLI). Research-backed, Vibe Check has been shown to nearly double AI agent success rates and halve harmful actions by introducing timely, reflective pauses. This plug-and-play server ensures LLMs remain aligned, reflective, and safe across complex and long-horizon agent workflows, acting as a crucial metacognitive layer.

주요 기능

• Session Constitution: Enforce per-session rules that the CPI layer will honor.
• History Continuity: Summarizes prior advice when `sessionId` is supplied, retaining context.
• Optional `vibe_learn`: Log mistakes and fixes for future reflection and self-improvement.
• 0 GitHub stars
• CPI Adaptive Interrupts: Phase-aware prompts that challenge assumptions and improve robustness.
• Multi-provider LLM Support: Compatible with Gemini, OpenAI, and OpenRouter.

사용 사례

• Preventing tunnel-vision, over-engineering, and reasoning lock-in in complex agent workflows.
• Enhancing reliability and safety for AI agents in ambiguous or high-risk tasks.
• Providing metacognitive oversight for AI agents involved in coding and development.