The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user prompts and codebase content to generate test fixtures in the evals/fixtures/ directory. While this processes external data, the activity is restricted to a specific local directory and follows a defined schema, posing minimal risk.
[PRIVILEGE_ESCALATION]: The skill implements a robust safety mechanism via the "§STOP" hard-stop list, which requires explicit user authorization before performing high-risk actions such as data deletion, production environment changes, or operations involving privacy exposure.

unknowns-discovery