DRAFT Protocol

About

DRAFT Protocol serves as an intake governance layer for AI tool calls, addressing the critical issue of AI agents acting on misinterpretations rather than true human intent. It employs a structured elicitation process, mapping every request across five dimensions—Define, Rules, Artifacts, Flex, Test—to surface assumptions and identify ambiguities. A mandatory confirmation gate blocks AI execution until all applicable fields are human-confirmed, preventing scope creep, wasted effort, and potential damage. DRAFT automatically classifies requests into three tiers of scrutiny (Casual, Standard, Consequential) and offers optional integration with any LLM for enhanced intelligence, making it a robust solution for ensuring AI safety and precision.

Key Features

• Broad compatibility with AI clients via MCP transports (stdio, SSE, Streamable HTTP) and a universal Chrome extension
• 0 GitHub stars
• Structured intent elicitation across five DRAFT dimensions (Define, Rules, Artifacts, Flex, Test)
• Adaptive scrutiny tiers (Casual, Standard, Consequential) with automatic classification and manual override
• Mandatory confirmation gate that blocks AI execution until human intent is explicitly verified
• Optional LLM integration for enhanced semantic classification, embedding-based field assessment, and context-aware suggestions

Use Cases

• Governing AI agent actions in sensitive environments, such as code generation, file management, or infrastructure deployment
• Preventing AI misunderstandings, scope creep, and wasted work in development and operational workflows
• Ensuring human intent is verified and confirmed before AI executes tasks across various AI chat platforms and tools