stress-field-surrogate
Warn
Audited by Gen Agent Trust Hub on Nov 7, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DYNAMIC_EXECUTION]: The script
scripts/stressd.pyperforms unsafe deserialization usingtorch.loadwithweights_only=Falsein thepredictfunction. - Evidence: The instruction
ck = torch.load(model, weights_only=False)allows the deserialization of arbitrary Python objects via thepicklemodule. - Impact: A maliciously crafted model file could execute arbitrary code when loaded by the skill.
- Mitigation: Use
weights_only=Trueand store metadata separately or ensure models only originate from verified sources. - [INDIRECT_PROMPT_INJECTION]: The skill ingests external data files without sufficient sanitization, presenting an attack surface for indirect prompt injection.
- Ingestion points: The skill reads
.npzdatasets and image files inscripts/stressd.pyvia thetrain,predict, andfemcommands. - Boundary markers: The skill lacks boundary markers or specific instructions to the agent to ignore embedded instructions in the processed data.
- Capability inventory: The skill has the ability to write to the file system using
torch.save,np.savez_compressed,np.save, andmatplotlib.pyplot.savefig. - Sanitization: There is no validation or sanitization of the input data content before it is processed or used in results.
- Mitigation: Implement strict schema validation for data files and use delimiters or safety instructions when the agent interacts with processed data.
Audit Metadata