location-world-producer
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external, potentially untrusted data to construct prompts for a 3D reconstruction model, creating a vulnerability surface.
- Ingestion points: The agent is instructed to read "Location production design context" and "Production Location Sheets" from the project environment to define the geometry and appearance of the generated 3D World.
- Boundary markers: No specific boundary markers or delimiters for the ingested content are defined in the instructions to prevent the model from misinterpreting data as instructions.
- Capability inventory: The skill possesses significant capabilities, including executing shell commands via the
renkuCLI, performing network requests to the World Labs API, and writing files to the project'stmp/directory. - Sanitization: The skill lacks explicit sanitization logic but implements a critical manual mitigation by requiring the agent to show the user the final prompt and image set for explicit approval before submission.
- [COMMAND_EXECUTION]: The skill relies on executing various subcommands of the
renkuCLI tool to manage project state, verify credentials, and initiate 3D model generation (e.g.,renku location world generate,renku asset select). These commands are scoped to the project environment.
Audit Metadata