The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It processes untrusted raw notes and shared context to extract keywords which are then passed to filesystem tools. The absence of boundary markers or instructions to ignore embedded commands within these inputs allows potentially malicious external data to influence filesystem search and read operations.
[DATA_EXFILTRATION]: The skill accesses a hardcoded local path (/Users/kp/Documents/Mario Jude 25) and generates public URLs for themariojude.com based on discovered filenames. This mechanism can lead to the exposure of private document titles and contents if the agent is manipulated into selecting sensitive files based on injected keywords.
[COMMAND_EXECUTION]: The skill utilizes filesystem tools (Filesystem:search_files and Filesystem:read_file) to interact with the host system. Because the search parameters are dynamically generated from user-provided themes without validation, it creates an exploitable surface for probing the local filesystem.

kmjp-social