The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions for the agent to perform a shallow clone of the official PhysicsNeMo repository from GitHub (https://github.com/NVIDIA/physicsnemo) if a local clone is unavailable. This download targets a trusted organization's repository and is restricted to read-only path discovery.
[COMMAND_EXECUTION]: The skill utilizes shell commands including ls and git clone, as well as specialized tools for file system search (Glob, Grep, Read). These capabilities are used to verify file paths and extract documentation strings from the repository to prevent hallucinations.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes data from external files within the repository (such as README.md files and source code docstrings) to generate its responses.
Ingestion points: Repository files located in physicsnemo/, examples/, and docs/ accessed via Read and Grep tools.
Boundary markers: The instructions do not specify explicit delimiters to isolate ingested content from the agent's internal reasoning.
Capability inventory: The agent has access to shell execution (Bash), repository management (git), and file discovery tools (Glob, Grep, Read).
Sanitization: There are no explicit sanitization routines defined for the data read from the repository.
[DATA_EXFILTRATION]: No evidence of unauthorized data exfiltration or suspicious network activity was found. Network access is constrained to the initial setup of the discovery environment from a known, trusted domain.

physicsnemo-discover