warp-debug-gradients
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute the user's simulation scripts to diagnose issues. While this involves running local code, it is the primary and intended purpose of a debugging skill. The instructions emphasize creating a 'shrunk reproduction' to minimize execution time and resource usage.
- [REMOTE_CODE_EXECUTION]: The skill mentions external documentation and examples (e.g.,
warp/examples/optim/example_fluid_checkpoint.py) but uses them as reference patterns rather than performing unverified remote downloads or executions. - [DATA_EXPOSURE]: Analysis of the references and instructions shows no attempts to access sensitive system files like credentials, SSH keys, or environment variables. The skill operates strictly on simulation data within the provided workspace.
- [PROMPT_INJECTION]: The instructions do not contain any patterns intended to bypass safety filters or override the agent's core instructions. It uses natural instructional language to guide the debugging process.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-provided simulation scripts, which is a data ingestion surface. However, the instructions mandate that the agent establish 'ground truth' via numeric verification (finite differences) rather than trusting claims in the code or comments, mitigating risk from malicious inputs.
Audit Metadata