The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data by analyzing user-provided reference videos to extract dialogue and beats, which can influence subsequent prompts as documented in SKILL.md and sections/05-two-ways-in.md.in.
Ingestion points: Reference videos are processed locally using ffmpeg and Whisper to extract content.
Boundary markers: The skill instructs the agent to "recreate the FORMAT, never the script," which helps mitigate the risk of following embedded instructions.
Capability inventory: The skill utilizes ffmpeg for local file manipulation, performs multiple network operations to the Novoads API, and writes files to the local outputs/ directory.
Sanitization: There is no explicit sanitization or filtering mentioned for the text extracted from reference videos via Whisper.
[COMMAND_EXECUTION]: The skill requires local execution of ffmpeg and jq for video analysis and assembly tasks, as well as shell scripts for environment verification (check-novoads-env.sh, setup.sh). These operations are essential to the skill's stated purpose and are documented as requirements.

claymation-ad