The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from video transcripts and metadata, creating an attack surface for indirect instructions.
Ingestion points: The workflow in SKILL.md uses get_media_context to retrieve voice-over transcripts and dialogue from user-provided compositions.
Boundary markers: The instructions do not specify any delimiters or safety prompts to ensure the agent ignores potential commands embedded within the retrieved transcripts.
Capability inventory: The skill is granted capabilities to modify composition timelines via apply_composition, adjust edit points using materializeGrid, and import external assets.
Sanitization: There is no evidence of sanitization or validation logic applied to the external transcript content before it is used to influence editing and mixing decisions.

add-music-bed