The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches the official PhysicsNeMo source code from NVIDIA's GitHub repository to enable live discovery of project artifacts and documentation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data by reading files such as READMEs, docstrings, and package initialization files from the target repository, which serves as a potential surface for indirect instructions.
Ingestion points: Read operations on examples/**/README.md, physicsnemo/**/*.py, and docs/*.rst within the cloned or local repository.
Boundary markers: Instructions explicitly require the agent to flag content from experimental/ directories as unstable.
Capability inventory: Access to Read, Glob, Grep, Bash, and git tools for repository interaction.
Sanitization: The skill emphasizes path verification using ls -d but does not define content-level sanitization for the retrieved documentation.
[COMMAND_EXECUTION]: The skill utilizes shell utilities including git clone for repository acquisition, as well as ls, grep, and glob for file discovery and path validation.
[REMOTE_CODE_EXECUTION]: The skill performs a clone of a remote Git repository; however, it includes strict constraints that the agent must never execute or import any of the downloaded code, using it exclusively for path discovery and documentation lookup.

physicsnemo-discover