nnsight-remote-interpretability
Pass
Audited by Gen Agent Trust Hub on Sep 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The documentation provides instructions for installing the
nnsightandtorchpackages from standard public package registries using pip. - [COMMAND_EXECUTION]: The skill includes standard shell commands for package installation and demonstrates how to set environment variables for API keys in a local terminal session.
- [INDIRECT_PROMPT_INJECTION]: The skill demonstrates processing arbitrary text prompts through neural network models. This represents a potential surface for indirect prompt injection where malicious instructions could be embedded in data processed by an agent using this skill.
- Ingestion points: Usage of
model.trace(prompt)andtracer.invoke(prompt)methods throughoutSKILL.mdandreferences/tutorials.mdto process external input. - Boundary markers: Absent; code examples treat input strings as raw text for the model without explicit delimiters or instructions to ignore embedded commands.
- Capability inventory: Local PyTorch activation manipulation and remote execution of computation graphs on the NDIF research platform.
- Sanitization: No explicit sanitization or validation of prompt content is demonstrated in the provided usage patterns.
Audit Metadata