The Agent Skills Directory

[SAFE]: The skill performs static code analysis based on internal documentation provided in the references/ directory. All operations are local to the agent's environment.
[DATA_EXFILTRATION]: The skill generates a report and saves it to a local directory (<appDataDir>/brain/<conversation-id>/). This is consistent with standard platform behavior for conversation artifacts and does not involve sending data to external domains.
[PROMPT_INJECTION]: The skill ingests untrusted code for review. While this is an entry point for potential indirect prompt injection, the skill does not possess high-risk capabilities like network access, command execution, or sensitive file access, which limits the impact of any embedded instructions in the analyzed code.

api-review