The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill transcribes audio and video files from potentially untrusted sources. The resulting text is intended for use in LLM contexts, creating an attack surface for indirect prompt injection where spoken instructions within the audio could influence the agent's behavior if the transcript is processed by the agent. \n
Ingestion points: Audio and video files (e.g., .mp3, .mp4) processed via the --input argument. \n
Boundary markers: None; the skill generates raw transcript files without delimiters or warnings for downstream LLMs. \n
Capability inventory: The skill has permissions for Bash, Write, Edit, Read, Grep, and Glob tools. \n
Sanitization: No sanitization or filtering of transcribed text is performed. \n- [COMMAND_EXECUTION]: The skill uses the Bash tool to perform media pre-processing tasks such as splitting, re-encoding, and extracting audio using ffmpeg and ffprobe as described in the documentation. \n- [DYNAMIC_EXECUTION]: The scripts/run.py file uses os.execv() to replace the current process with a Python interpreter from a local virtual environment (.runners-venv). This is a standard mechanism to ensure the skill runs in a controlled environment. \n- [EXTERNAL_DOWNLOADS]: The documentation suggests the installation of the pysrt Python package via pip to enable advanced subtitle manipulation and timestamp shifting.

transcribe-maker