The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses npx hyperframes to download and execute the necessary tools for transcription and video rendering from the npm registry. This is part of the core operational design.
[COMMAND_EXECUTION]: The workflow involves executing local system commands including ffmpeg, ffprobe, mkdir, cp, and sed to extract audio, gather media metadata, and manage the project directory.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes transcript.json files (derived from user audio) to guide the agent in generating graphic card designs. This creates an indirect injection surface if the input audio contains malicious instructions intended to influence the agent's code generation.
Ingestion points: Transcript data is ingested from transcript.json in Step 5.
Boundary markers: No specific delimiters or "ignore instructions" warnings are defined for the transcript ingestion phase.
Capability inventory: The skill has the capability to generate HTML/JavaScript and trigger a local rendering process via npx hyperframes.
Sanitization: The skill enforces a strict HTML contract (forbidding scripts and external URLs in card fragments) which is validated by a linter in the hyperframes CLI.
[DYNAMIC_EXECUTION]: The agent dynamically generates GSAP animation code and HTML structures at runtime based on the processed transcript. This code is then saved to local files and executed by a headless browser during the video rendering step.

hyperframes-talking-head-recut