The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes multiple system commands using ffmpeg for audio extraction and subtitle embedding, npx ts-node for running TypeScript scripts, and open to display the final output.
[EXTERNAL_DOWNLOADS]: The pipeline triggers npm install within the directories of its dependent skills (/transcribe and /embed-subtitles) to ensure the necessary environment for its Node.js scripts.
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) because it ingests and processes untrusted data extracted from video files.
Ingestion points: Transcription text and SRT files derived from user-provided video files are read by the LLM for translation and semantic refinement in Step 2 and Step 3.
Boundary markers: Absent. There are no explicit delimiters or instructions provided to the agent to disregard potential instructions embedded within the transcript content.
Capability inventory: The skill has the capability to execute shell commands (ffmpeg, npx) and interact with the file system based on paths derived from the input video.
Sanitization: No sanitization or validation of the transcribed text is performed before it is re-processed by the LLM or used as input for the subtitle embedding step.

translate-video