The Agent Skills Directory

[PROMPT_INJECTION]: The instructions are strictly technical and focused on API usage. No patterns matching safety filter bypasses or agent instruction overrides were detected.
[DATA_EXPOSURE_AND_EXFILTRATION]: No sensitive file paths, hardcoded credentials, or network exfiltration patterns were found. The skill operates on local or provided USD scene data using standard tools.
[OBFUSCATION]: Analysis for Base64, hex encoding, zero-width characters, and homoglyphs returned no findings. The content is presented in clear, plain text.
[REMOTE_CODE_EXECUTION]: No remote script downloads or piped execution patterns (e.g., curl|bash) were detected. The skill references local code snippets for implementation.
[INDIRECT_PROMPT_INJECTION]: The skill describes ingesting external USD data through open_usd and open_usd_from_string in SKILL.md. While these are ingestion points for untrusted data, the risk is minimal due to the specialized nature of the USD format and the restricted toolset (Read, Grep) available to the agent. No sanitization or boundary markers are specified, but the capabilities are not highly exploitable.

runtime-loop