AgentGuard FAQs

Question 1

What problem does AgentGuard solve for AI agents?

Accepted Answer

AgentGuard addresses the 'false progress' problem, where AI agents report task completion without actual execution. It independently verifies agent outputs against predefined contracts and evidence requirements, ensuring true accountability and reliable task execution.

Question 2

How does AgentGuard verify an agent's work?

Accepted Answer

You define a 'contract' outlining the task and specific evidence required for completion. Agents submit this evidence (e.g., command outputs, HTTP checks, file content), which AgentGuard then independently verifies before marking the task as genuinely done.

Question 3

What is 'MCP' and how does AgentGuard utilize it?

Accepted Answer

MCP stands for Model Context Protocol. AgentGuard functions as an MCP server, providing a standardized JSON-RPC 2.0 interface (via stdio or HTTP Server-Sent Events) for agents to interact with the system, receive tasks, submit evidence, and manage their operations securely and efficiently.

Question 4

What is the purpose of AgentGuard's audit log?

Accepted Answer

AgentGuard maintains an immutable, append-only audit log that records every state transition and verification event. This log is crucial for compliance, providing complete transparency, and facilitating debugging by offering a detailed history of all agent activities and contract fulfillments.

Question 5

Can AgentGuard manage multi-agent workflows and pipelines?

Accepted Answer

Yes, AgentGuard offers 'Pipeline Mode' to orchestrate multi-agent workflows as Directed Acyclic Graphs (DAGs) with artifact passing. It also features 'Free Mode,' enabling orchestrator agents to decompose high-level goals into sub-contracts for worker agents with enforced limits.

AgentGuard

AgentGuard

Características Principales

Casos de Uso

Características Principales

Casos de Uso