Video Analyzer
SuspiciousAudited by ClawScan on May 10, 2026.
Overview
The skill mostly matches its video transcription purpose, but its script builds shell commands from user-controlled input in a way that could run unintended local commands.
Review carefully before installing. The skill appears intended for local video transcription, but it should be fixed to validate parameters and avoid shell=True; until then, only use ordinary trusted video URLs and standard language codes.
Findings (2)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
A malicious or accidental language parameter could make the agent run commands outside the intended video transcription task.
The language argument is not restricted to a safe pattern and is interpolated into commands executed with shell=True. A crafted language value containing shell metacharacters could execute commands as the local user.
result = subprocess.run(cmd, shell=True, check=check, capture_output=True, text=True)
...
parser.add_argument("--lang", default="en", help="Language code (e.g., 'en', 'it')")
...
lang_flag = f"-l {lang}" if lang != 'en' else ""Avoid shell=True, pass subprocess arguments as lists, validate language codes with a strict allowlist or regex, and quote every user-controlled value with a safe library if a shell is unavoidable.
The first transcription may download a large model file from the internet and store it in the Homebrew share directory.
The skill downloads Whisper model files from Hugging Face at runtime without a checksum or signature check. This is purpose-aligned, but it relies on an external artifact remaining trustworthy.
WHISPER_BASE_URL = "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/"
...
run_cmd(f'curl -L "{url}" -o "{model_path}"')Pin expected model hashes and verify downloads before use, or clearly document the exact model source and integrity expectations.