Ensures AI agents perform tasks as contracted by independently verifying required evidence.
AgentGuard addresses the critical 'false progress' problem in AI agent operations, where agents might report task completion without actual work being done. It acts as an MCP server that sits between users and their AI agents, enforcing accountability through predefined contracts. Users specify tasks and required evidence, which AgentGuard independently verifies, preventing blind trust in agent outputs and ensuring that 'no evidence' truly means 'not done'. Every state transition and verification is recorded in an append-only audit log, providing a complete and immutable record of agent activities.
