The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill contains a persistence feature that writes documentation to the local filesystem. This capability was audited for path traversal risks. The implementation uses a "safe_slug" utility in "design_system.py" which strictly enforces an alphanumeric character set with hyphens and underscores for all user-controllable path segments (project names and page names). This design effectively prevents directory traversal attacks and unauthorized file system access.
[REMOTE_CODE_EXECUTION]: User-provided queries are passed to a local search engine based on the BM25 algorithm. Analysis of the "core.py" script confirms that queries are sanitized using regular expressions ("[^\w\s]") to remove non-word characters before processing. This prevents the input from being used to manipulate the search logic or execute arbitrary commands.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of search queries and project titles and interpolates them into generated markdown files. While this creates a vulnerability surface, the risk is mitigated because the output is restricted to documentation files and the skill lacks any automated capabilities to execute the generated content or perform network exfiltration.

ui-ux-pro-max