S03 · Integration security

Secrets, dependencies and execution boundaries

Leaked credentials or unsafe dependencies turn agent execution into a path to wider systems.

Requirement

Keep secrets outside prompts and code, scan dependencies and isolate generated commands or code with explicit resource limits.

Expected evidence

  • Secret-management and rotation records.
  • Dependency inventory and sandbox policy.

Proposed verification

  1. Inspect prompts, logs and repositories for exposed secrets.
  2. Run a benign escape and resource-exhaustion test in the authorized sandbox.

Limit

A clean scan is point-in-time and does not cover unknown vulnerabilities.

Sources