Can you reliably audit what a nested automation agent does inside your infrastructure?
Session recording is the only way to guarantee a complete audit of those actions.
Today many teams let CI/CD runners, AI‑driven assistants, or service‑account bots connect directly to databases, SSH hosts, or Kubernetes clusters. Those agents often inherit a static service credential that is shared across pipelines. Because the connection bypasses any visible gateway, there is no built‑in record of the commands issued, the queries run, or the data returned. If an agent is compromised or a misconfiguration slips through, the activity disappears into the logs of the target system, making forensic analysis painful.
Even when an organization enforces least‑privilege roles and OIDC tokens for the agents, the request still travels straight to the target. The token proves who may start the session, but it does not capture what the agent actually does once the tunnel is open. Without a dedicated recording layer, you lack command‑level audit, replay capability, and the ability to mask sensitive fields in responses.
hoop.dev solves this gap by inserting a Layer 7 gateway between the agent’s identity and the infrastructure resource. The gateway becomes the sole data‑path component that can observe, record, and optionally transform every protocol exchange. Because the proxy holds the credential, the nested agent never sees the secret directly, and every session is captured for replay and audit.
Why the data path matters
The authentication step – OIDC or SAML verification, role assignment, and token issuance – decides who may start a connection. That is the setup phase. It is necessary for access control but it does not enforce any runtime guardrails. Only when the request passes through the gateway can enforcement be applied. By positioning hoop.dev as the identity‑aware proxy, you guarantee that every byte of traffic is subject to policy before it reaches the target.
Enforcement outcomes for nested agents
- hoop.dev records each session at the protocol level, preserving a faithful replay that includes timestamps, commands, and responses.
- hoop.dev masks sensitive fields in real time, ensuring that secrets or personal data never appear in logs or recordings.
- hoop.dev can require just‑in‑time approval for high‑risk commands, pausing execution until a human reviewer signs off.
- hoop.dev blocks disallowed operations before they reach the backend, reducing blast radius from a compromised agent.
All of these outcomes exist only because the gateway sits in the data path. Remove the gateway and the same agent would once again act without visibility or control.
How nested agents interact with the gateway
When a nested agent starts, it obtains an OIDC token from the organization’s identity provider. The token is presented to the gateway, which validates it and extracts group membership. The agent then opens a standard client connection – for example psql for PostgreSQL or ssh for a host – but the network address points to the gateway instead of the database directly. The gateway forwards the traffic to the internal resource, applying the configured policies on the fly.
This pattern works for any supported target, including databases, Kubernetes exec sessions, and remote desktop connections. Because the gateway is protocol‑aware, it can parse commands, apply inline masking, and write a complete audit trail without requiring any changes to the agent’s code.
Getting started
Deploy the gateway using the Docker Compose quick‑start or the Kubernetes manifest described in the getting‑started guide. Register each nested agent’s target as a connection, configure the credential that the gateway will use, and enable session recording in the policy definition. Detailed feature explanations are available in the learn section. The open‑source repository on GitHub contains the full source code and example configurations.
Explore the hoop.dev source on GitHub to see how the proxy is built and to contribute improvements.
FAQ
How does session recording handle high‑throughput workloads?
Recording is performed as a stream that the gateway can scale horizontally, ensuring that concurrent sessions do not introduce noticeable latency.
Can I retain recordings for compliance periods longer than a year?
Yes. The gateway produces a durable audit record for each session, and you can store those records in any long‑term archive that meets your retention policy.
Will masking affect the correctness of the underlying operation?
No. Masking is applied only to data that is sent back to the client or stored in logs. The original response is forwarded unchanged to the target system, preserving functional behavior.
