The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external text from SRT files and media data (video/audio) to drive visual effects. Maliciously crafted input could attempt to influence the agent's behavior during the analysis or script-building phases.
Ingestion points: The parse_srt function in references/inputs.md reads content from user-supplied subtitle files. Media analysis routines ingest raw audio and video data for feature extraction.
Boundary markers: There are no specific delimiters or protective instructions surrounding the ingested text data in the current architecture.
Capability inventory: The generated scripts utilize subprocess.run to call ffmpeg and ffprobe, and perform network requests to the ElevenLabs API via the requests library.
Sanitization: Content extracted from external files is processed without explicit sanitization or filtering against instructional patterns before being interpolated into the rendering logic.
[DYNAMIC_EXECUTION]: The skill's primary workflow involves the AI agent generating a standalone Python script which is then executed to perform the media processing and video rendering.
Evidence: Instructions in SKILL.md guide the agent to 'Build the Script,' and the README.md notes that the agent 'writes the renderer, runs it.' The architecture relies on runtime code generation to customize the visual output based on user requests.

ascii-video