The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions frequently utilize npx hyperframes to run transcription, environment diagnostic, and video rendering tasks. This pattern executes code directly from the NPM registry. It also includes an update command, npx hyperframes skills update talking-head-recut, to refresh skill-related code.
[DYNAMIC_EXECUTION]: The skill designs and generates HTML/CSS fragments for graphic cards which are then assembled into a master composition file. This generated content is executed within a headless browser context during the rendering phase via the hyperframes render command. The skill specifies strict constraints for this generated code, such as prohibiting <script> tags and external URLs to mitigate risks.
[INDIRECT_PROMPT_INJECTION]: The skill has a data ingestion surface where it processes video transcripts to influence the content and timing of graphic overlays.
Ingestion points: The transcript.json file, which contains word-level ASR data derived from the user's video audio, is read by the agent to drive card design.
Boundary markers: Absent; the skill relies on the agent to interpret a flat array of word objects without formal delimiters for instructions versus data.
Capability inventory: Shell command execution (ffmpeg, npx hyperframes) and local file system writes.
Sanitization: Absent; text from the transcript is interpolated directly into generated HTML cards.
[COMMAND_EXECUTION]: The skill executes multiple shell commands using ffmpeg and ffprobe to extract audio, retrieve media metadata, and re-encode video files with specific keyframe intervals for the renderer.

talking-head-recut