Skillv1.0.0

ClawScan security

Bilibili Video Transcriber · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

BenignApr 15, 2026, 1:25 AM

Verdict: Benign
Confidence: medium
Model: gpt-5-mini
Summary: The package and runtime instructions are coherent with a Bilibili video transcription tool — the code, CLI, and docs match the described purpose — but it handles sensitive user cookies and references a non-official mirror, so review before use.
Guidance: This skill appears to be what it claims: a Bilibili video transcriber that downloads audio and runs a local Whisper model. Before installing or running it, consider: 1) The tool expects you to supply Bilibili cookies (SESSDATA, bili_jct, etc.) — these are sensitive session credentials. Only paste/store cookies you trust and keep the cookie file permissions restricted. 2) The code and docs reference a custom mirror (https://hf-mirror.com) for model downloads — verify you trust that mirror before allowing model downloads from it (or configure an official Hugging Face endpoint). 3) The installer will create config and cookie files in your home directory and may create a symlink under ~/.local/bin; review setup.py if you want different locations. 4) Notification options (SMTP/webhook) exist in the config; do not populate them with real credentials or endpoints unless you trust the code and destination. 5) If you need higher assurance, review the included Python files (bilibili_transcriber.py, cli.py, setup.py) yourself or run them in a sandboxed environment; consider using a throwaway Bilibili account or API token rather than your main account if possible.
Findings: [pre-scan-injection-signals] expected: No pre-scan injection signals detected. Absence of regex matches is not an assurance of safety; here the repository includes many plain-text scripts and no obfuscated or clearly malicious patterns.

Review Dimensions

Purpose & Capability: okName and description (B站视频转录、字幕下载/分析) align with the included code, CLI, and dependencies (bilibili-api, faster-whisper, pydub, requests). Required binaries (python3 or ffmpeg) are reasonable for media download/processing and local Whisper usage.
Instruction Scope: noteSKILL.md and README instruct the agent/user to read/write a Bilibili cookie file and to use environment variables (BILIBILI_COOKIE, HF_ENDPOINT, HTTP_PROXY) and local paths (~/.bilibili_cookie.txt, ~/.config/...). Those actions are expected for authenticated Bilibili API access and model/mirror configuration, but the runtime instructions do involve handling sensitive credentials (browser cookies) and creating config files in the user's home directory.
Install Mechanism: okThere is no remote download/install spec in the registry metadata (instruction-only skill). The repository contains normal packaging and setup scripts that install Python dependencies via pip. No evidence of downloads from untrusted shorteners or personal IPs; the only non-official host referenced is a domestic mirror (https://hf-mirror.com) used for Hugging Face endpoints.
Credentials: noteThe skill does not declare required environment variables in the registry metadata, but the code and docs use BILIBILI_COOKIE and allow exporting HF_ENDPOINT/HTTP_PROXY. Asking for Bilibili cookies is proportionate to the stated functionality, but these are sensitive credentials and the mismatch between declared requirements (none) and the actual use of env vars should be highlighted to users.
Persistence & Privilege: okThe package creates config files and may create a cookie file template, symlinks (~/.local/bin) and output directories under the user's home — standard for CLI tools. always: false and normal autonomous invocation settings. No code attempts to modify other skills or system-wide agent settings.