The Agent Skills Directory

[SAFE]: No malicious patterns were identified. The skill follows best practices for media editing, such as requiring transcription before performing cuts and providing human-review checkpoints for the final output.
[INDIRECT_PROMPT_INJECTION]: The skill processes speech transcripts derived from external media files, which represents a potential attack surface for indirect prompt injection.
Ingestion points: The skill utilizes the transcribe tool and the semanticProfile.transcript object to ingest speech data from clips.
Boundary markers: The instructions include explicit constraints to prevent accidental obedience to fragmented words and require that any text used in graphical inserts must match the actual spoken transcript.
Capability inventory: The tools available to the skill are restricted to media manipulation, including cutRanges, addCaptions, designInserts, fetchStockMedia, and setAudio for volume leveling.
Sanitization: The skill enforces strict validation, such as the wordsMustShareAWord rule in references/inserts.md, ensuring that the model does not generate content that was not explicitly stated in the source audio.

talking-head