The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The pipeline utilizes the bunx runner to fetch the hyperframes tool from the npm registry and relies on the Deepgram CLI for speech synthesis and recognition services.
[COMMAND_EXECUTION]: Orchestrates the use of ffmpeg and ffprobe for media file manipulation, such as audio encoding and duration probing. Commands are executed through execFileSync with properly structured arguments to avoid shell injection.
[PROMPT_INJECTION]: The skill handles untrusted Markdown input as a core part of its workflow. It mitigates indirect prompt injection risks by explicitly instructing the agent to perform an 'adversarial faithfulness audit' to ensure the distilled narration remains grounded in the source text and free from hallucinations.

humanlayer-to-video