agent-runtime-gateway-smoke-test
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions involve executing shell commands (specifically
curl) to perform health checks on local and remote API endpoints. It also directs the agent to manage process lifetimes (starting and stopping APIs and tunnels) and modify application configurations (changing web app routes). - [INDIRECT_PROMPT_INJECTION]: The skill creates a diagnostic loop where a remote sandbox performs a callback to a local API. This represents a potential surface for indirect prompt injection if the remote sandbox processes untrusted data that is then ingested by the local API.
- Ingestion points: Local API logs and gateway route responses received during the callback phase.
- Boundary markers: The skill suggests validating the output against an "exact expected answer," providing a basic form of output verification, though it lacks explicit delimiters for the incoming callback data.
- Capability inventory: The agent has the capability to perform network requests (
curl) and manage local system processes. - Sanitization: The instructions include strong mandates for sanitization, specifically requiring that credentials, personal paths, and raw logs be excluded from reports. It further references
hermes-importsas a tool for sanitizing workflows.
Audit Metadata