PromptSpeak

Pre-execution governance for AI agents with 9-stage validation, drift detection, human approval queues, and circuit breakers.

PromptSpeak intercepts AI agent tool calls before execution and validates them through a 9-stage governance pipeline covering symbolic frame validation, behavioral drift detection, vulnerability scanning for 10 security patterns, and human-in-the-loop approval queues. It exposes 56 MCP tools across governance, hold management, agent lifecycle control, and delegation domains with sub-millisecond validation latency. The server is installable via npm and includes 829 passing tests.