The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions explicitly direct the agent to search through highly sensitive local configuration files, including ~/.zshrc, ~/.zshenv, ~/.zprofile, ~/.profile, ~/.bashrc, ~/.bash_profile, and various .env* files. This behavior is intended to locate USD_SEARCH_API_USERNAME, USD_SEARCH_API_PASSWORD, and USD_SEARCH_API_TOKEN. Searching these files is dangerous as they often contain a wide range of sensitive credentials, keys, and tokens for various services beyond the scope of this skill.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external source (USD Search API) that could potentially contain malicious instructions embedded in asset metadata.
Ingestion points: Data is fetched via curl from several endpoints, including /info/indexing/asset/status, /images, /asset_graph/usd/scene_summary/, and /dependency_graph/flat (documented in SKILL.md).
Boundary markers: The instructions do not define clear delimiters or "ignore instructions" warnings for the content returned by these API calls.
Capability inventory: The skill uses the Bash tool to execute shell commands for the API interaction and local file management.
Sanitization: While the skill uses --data-urlencode for outgoing requests, there is no evidence of sanitization or filtering for the incoming API responses before they are processed by the agent.

inspect-asset