The Agent Skills Directory

[PROMPT_INJECTION]: Potential surface for indirect prompt injection discovered in the handling of external scan data.\n- Ingestion points: The skill instructions involve analyzing untrusted external content including vulnerability scan results, plugin descriptions, advisories, and service banners (SKILL.md).\n- Boundary markers: No specific technical delimiters (such as XML tags or unique string markers) are defined to isolate untrusted data inputs.\n- Capability inventory: The agent has access to file system tools (Read, Grep, Glob) which could be misused if an injection is successful.\n- Sanitization: The skill explicitly includes a 'Prompt Injection Safety Notice' that instructs the agent to ignore any commands embedded in target data or banners and to report them as suspicious.

scanner-tuning