The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of the subprocess module to call ffmpeg and ffprobe for media processing tasks such as audio extraction, duration calculation, and video muxing.
Evidence: Found in scripts/lib.py (function run_cmd), scripts/dub.py (_ffmpeg_extract_wav, _cut_wav, _time_fit, _mux), and scripts/voiceover.py (_prepare_voice_reference).
Note: While the use of list-based arguments mitigates shell injection risks, the ability to execute shell-level commands remains a significant capability.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes text from external JSON files and interpolates it into payloads sent to remote synthesis and recognition services.
Ingestion points: The skill reads from narration.json (in voiceover.py) and dub_script.json (in dub.py).
Boundary markers: No explicit boundary markers or "ignore" instructions are used to wrap the untrusted text before it is sent to the TTS/ASR APIs.
Capability inventory: The skill possesses network access (API calls to MiMo and Fish Audio), file system write access (saving audio segments), and shell execution capabilities (ffmpeg).
Sanitization: A _clean_narration_text function in voiceover.py removes markdown formatting, emojis, and specific annotations, but it is not designed to detect or sanitize adversarial prompt instructions.

video-voiceover