Skill flagged — suspicious patterns detected
ClawHub Security flagged this skill as suspicious. Review the scan results before using.
subtitle-extractor
v1.0.0Subtitle extractor for Bilibili, YouTube, Xiaohongshu, Douyin, and local files. Extracts native subtitles or Whisper transcription in original format. Agent...
⭐ 1· 54·0 current·0 all-time
by@poltawa
MIT-0
Download zip
LicenseMIT-0 · Free to use, modify, and redistribute. No attribution required.
Security Scan
OpenClaw
Suspicious
medium confidencePurpose & Capability
The name/description match the code: it uses yt-dlp for native subtitles and faster-whisper/ffmpeg for transcription. However the registry metadata only lists yt-dlp as a required binary while the SKILL.md and script clearly require ffmpeg (for transcription) and the Python faster-whisper package. That mismatch between declared requirements and actual runtime needs is inconsistent.
Instruction Scope
SKILL.md forcefully requires running Steps 1–4 before answering user requests (non-negotiable), which can trigger network access, model downloads, and reading local cookie files even when not necessary. The SKILL.md claims cookie files are "read locally only, never transmitted externally," but the script passes cookies to yt-dlp for network requests (so cookie data will be sent to remote services). The script also auto-discovers any *.txt with 'bilibili' in its filename in the skill directory, which could pick up unintended cookie files.
Install Mechanism
There is no install spec (instruction-only), but the included Python script performs runtime downloads (Whisper models via faster-whisper/HuggingFace or ModelScope through modelscope), which will fetch ~100s of MBs of model data and write them to disk. Runtime downloads are not flagged in a manifest and may require network access and storage.
Credentials
The registry declares no required env vars, but SKILL.md and the script read/use VIDEO_SUMMARY_WHISPER_MODEL and reference VIDEO_SUMMARY_COOKIES. These env vars can change behavior (point to local models or cookie paths). The skill reads files from its directory and may write model files and outputs under the skill folder; the use of env vars and file discovery is not declared in the metadata.
Persistence & Privilege
The skill is not marked always:true and does not request platformwide privileges. It writes models and outputs to its own skill directory and does not modify other skills or system-wide agent settings. Autonomous invocation is allowed (platform default) but not combined with 'always'.
What to consider before installing
This skill appears to implement subtitle extraction and optional Whisper transcription, but review these issues before installing:
- Cookie handling: the SKILL.md claim that cookies are "never transmitted externally" is incorrect — cookies passed to yt-dlp are used for network requests and will reach the video service. Do not place sensitive cookies in the skill directory unless you understand the exposure.
- Mandatory pre-run: the skill requires Steps 1–4 to run before responding. That can trigger network access and large model downloads even if the user's request doesn't need them. If you want to avoid unexpected network activity, run in an isolated environment or require explicit user approval before extraction/transcription.
- Missing/unclear requirements: ffmpeg and the faster-whisper Python package (and optionally modelscope) are required for transcription flows but are not listed in the registry metadata. Ensure you install/verify these tools first.
- Runtime downloads: transcription may auto-download Whisper models (~150MB for base); this consumes bandwidth and disk. Consider setting VIDEO_SUMMARY_WHISPER_MODEL to a vetted local model directory to avoid automatic downloads.
- File discovery: the script auto-discovers any *.txt with 'bilibili' in the skill directory. Keep that directory free of unrelated cookie files.
If these behaviors are acceptable, proceed; otherwise request the author clarify the manifest (declare ffmpeg/faster-whisper env vars) and remove the misleading security statement about cookie transmission.Like a lobster shell, security has layers — review code before you run it.
latestvk972h15cf2qz2rfber6d1e44ph83jvyv
License
MIT-0
Free to use, modify, and redistribute. No attribution required.
Runtime requirements
Binsyt-dlp