The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing of user-supplied text prompts within a framework that possesses network capabilities for remote model execution.
Ingestion points: The trace() method and various prompt variables in SKILL.md and references/tutorials.md ingest external text data into the agent's context.
Boundary markers: The provided code examples do not implement delimiters or safety warnings to ignore instructions that could be embedded within the analyzed prompts.
Capability inventory: The skill utilizes the nnsight library's capability to perform network operations for remote model execution on the NDIF (National Deep Inference Facility) infrastructure.
Sanitization: No input sanitization or validation of the processed prompts is implemented in the provided workflows.
[EXTERNAL_DOWNLOADS]: The skill provides instructions to download and use external Python packages and utilizes a remote research service for computation.
Libraries: Instructions to install nnsight, torch, and the vllm extension from public registries.
Remote Infrastructure: The skill references the ndif.us service and ndif-team GitHub repositories for executing computation graphs on large-scale models.

nnsight-remote-interpretability