The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill exhibits an attack surface for indirect prompt injection by processing untrusted data from external and internal sources.\n- Ingestion points: Data enters the agent context through WebSearch (Step 1), the validated_ideas.json file (Before You Start), and direct user input.\n- Boundary markers: Absent. The instructions do not define delimiters or provide 'ignore embedded instructions' warnings for the data being processed.\n- Capability inventory: The skill utilizes WebSearch and possesses file-write capabilities (video-brief-{slug}.md).\n- Sanitization: There is no evidence of sanitization, escaping, or schema validation for the data ingested from the web or local files.

yt-brief