The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes a robust safety notice that explicitly warns the agent to treat reviewed content as data and not to follow instructions found within it. This effectively mitigates potential indirect prompt injection attacks from the files being analyzed.
[COMMAND_EXECUTION]: The toolset is restricted to Read, Grep, and Glob, which are used appropriately for static analysis of code and configuration files. No dangerous system-level commands or unauthorized tool usages were detected.
[DATA_EXFILTRATION]: The skill performs scans for PII and credentials as part of its auditing function. These operations are local and read-only, with no network-based transmission of the discovered information.

ai-data-privacy