search-in-scene
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill instructs the agent to read sensitive local configuration files (~/.bashrc, ~/.zshrc, ~/.profile, etc.) to identify environment variable definitions for API authentication. While targeted at specific variable names, this pattern could be used to expose credentials if the API URL is directed to an untrusted source.
- [COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute curl commands, passing authentication headers (Basic or Bearer tokens) to interact with remote spatial query endpoints.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted JSON data from external API endpoints (USD Search API). Ingestion points: Data enters the agent context through JSON responses from endpoints like /scene_summary and /prims. Boundary markers: The instructions lack explicit delimiters or warnings to ignore potentially malicious content embedded in the API responses. Capability inventory: The agent has access to shell execution (curl) and local file reading. Sanitization: There is no evidence of validation or sanitization of the API data before it is interpreted by the agent.
Audit Metadata