Integrating target agents as Hook
AgentSuite-Blue's Prompt Guard and Action Guard can be integrated into the target agent as hooks. Prompt Guard runs as a pre-agent and post-agent hook to inspect every incoming prompt and agent response, while Action Guard runs as a pre-tool and post-tool hook to inspect every tool call.
Both components also report to our dynamic observability service, giving you end-to-end visibility into the agent's run-time trajectory. The figure below illustrates the architecture.
Note: The static protections (MCP Guard, Skill Guard, and Static Observability) do not need to be integrated into the target agent — they can be accessed directly from our dashboard. We also expose them as standalone endpoints that can be called without going through the dashboard (e.g., for CI/CD integration). See the MCP Guard SDK for an example.
AgentSuite-SDK & Hook
Writing hooks relies on our AgentSuite-SDK, which wraps Prompt Guard, Action Guard, and dynamic observability as ready-to-use endpoints.
Integration Guide
AgentSuite-Blue integrates easily with a wide range of agent SDKs and well-known coding agents. The list below covers the currently supported targets. We use Action Guard to demonstrate the connection.