The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in several scripts (scripts/anibon/ytdlp.py, scripts/benchmark_check.py, scripts/detect_hallucinations.py) to execute command-line tools such as yt-dlp, ffmpeg, and whisper-cli. While these support the skill's primary purpose, executing commands with variable arguments increases the attack surface.
[REMOTE_CODE_EXECUTION]: In scripts/detect_hallucinations.py, the skill dynamically constructs a path to a script located in the user's home directory (~/.agents/skills/whisper-corruption-recovery/scripts/fix_hallucinations.py) and executes it. This creates a dependency on external code that is not part of the skill's own package.
[COMMAND_EXECUTION]: The script scripts/anibon/markdown.py utilizes the __import__ function to dynamically load the sys module. Dynamic module loading can be used to bypass static analysis and is considered an obfuscation technique.
[PROMPT_INJECTION]: The skill processes untrusted text data ingested from YouTube transcripts and Whisper transcription outputs. It lacks boundary markers or explicit instructions to ignore embedded commands, making it susceptible to indirect prompt injection if the processed content contains malicious instructions designed to influence the agent's behavior.
Ingestion points: scripts/anibon/ytdlp.py (YouTube transcripts), scripts/anibon/chunk_io.py (JSON/XML chunk files), scripts/detect_hallucinations.py (Whisper JSON output).
Boundary markers: None identified in the text processing pipeline.
Capability inventory: File writes, command execution via subprocess.run (yt-dlp, ffmpeg, whisper-cli, fix_hallucinations.py).
Sanitization: scripts/anibon/cleaner.py performs regex-based text cleaning for quality, but does not provide security-focused sanitization.

anibon-local-transcription