The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (Tweets, articles, and video links) to generate YouTube content, creating an attack surface for indirect prompt injection.\n- Ingestion points: Content enters the context via external links or text from 'Tweet / X post', 'Link to someone else's YouTube video', and 'Article or changelog' as defined in the Step 1 workflow of SKILL.md.\n- Boundary markers: Absent. The instructions do not define delimiters to encapsulate the external content or explicitly instruct the model to ignore instructions contained within the ingested text.\n- Capability inventory: The skill primarily generates text (titles, hooks, and video structures) and does not exhibit high-risk capabilities like shell execution, network exfiltration, or file-system modification in the provided files.\n- Sanitization: Absent. No validation or filtering logic is present to sanitize the ingested content before it is processed by the model.

youtube-producer