The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external USD, USDA, and USDZ files which could potentially contain malicious instructions embedded within metadata, comments, or property values.
Ingestion points: Commands such as usdcat, usdtree, and usdedit mentioned in SKILL.md and their respective reference files (references/usdcat.md, references/usdtree.md, references/usdedit.md) read data from external files into the agent's context.
Boundary markers: The instructions do not specify explicit delimiters or "ignore embedded instructions" warnings for the content of the USD files being processed.
Capability inventory: The skill documented across all reference files utilizes several subprocess calls (usdcat, usdchecker, usdrecord, usdtree, usdzip, usdedit) to operate on the ingested files.
Sanitization: There is no evidence of content sanitization or validation to prevent the interpretation of textual data within USD files as instructions.
[COMMAND_EXECUTION]: The skill documentation encourages the use of multiple command-line utilities to inspect, convert, and validate assets.
Evidence: references/command-line-tools.md and other reference files provide examples for executing usdcat, usdchecker, usdrecord, usdtree, usdzip, and usdedit with various arguments, which involves shell command execution based on the provided file paths.

usd-editor