The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes narration scripts resolved from user input or AI-generated drafts, which provides a surface for indirect prompt injection where malicious instructions could be embedded in the text intended for synthesis.\n- Ingestion points: Input for scripts is gathered from user text, project data, and attachments throughout the workflow in SKILL.md.\n- Boundary markers: The instructions do not define the use of delimiters or markers to isolate the script content from the agent's instructions during synthesis.\n- Capability inventory: The skill utilizes submit_avatar_video for content generation and manage_avatar for managing digital identities.\n- Sanitization: Instructions explicitly require preserving the user's wording and punctuation, which avoids modification but also omits sanitization or filtering of the input script.

digital-human