CZX2244/dsh-bilibili
Bilibili extraction tool plugin: give the agent a Bilibili link and it extracts metadata, a complete timestamped transcript (auto-transcribed via Bijian ASR by default when no subtitle track exists, or local sherpa-onnx/whisper.cpp fully offline), hot comments with replies, and danmaku with density-peak timeline samples. Optional frame vision descriptions for vision-less main models via local Ollama/llama.cpp (Qwen3-VL 2B/4B/8B) or any OpenAI-compatible vision API. Automatic scene-change frame selection with sharp-frame preference (FFmpeg blurdetect), two-pass workflow (transcript first, then explicit [建议抓帧] mm:ss frame requests), replaceable summary template, download-first capture with remote fallback, and exponential backoff on Bilibili 412 rate limits.
Install
dsh plugin --profile web add git+https://github.com/CZX2244/dsh-bilibiliGit URL install per README (English edition): dsh plugin --profile web add git+https://github.com/CZX2244/dsh-bilibili or dsh plugin --profile web add ./dsh-bilibili from a clone. No npm package. Bundles no third-party binaries or models; invoked open-source projects listed in THIRD_PARTY_NOTICES.md. MIT.
Compatibility
DeepSeek Harness tool plugin. Extracts via public Bilibili endpoints; optional local ASR engines (sherpa-onnx, whisper.cpp) and vision backends (Ollama/llama.cpp, OpenAI-compatible) for offline operation.
Details
- Repo: CZX2244/dsh-bilibili
- Category: Other
- Stars: 3
- Version: GitHub source (no npm package)
- Last push: 2026-08-15
- First seen: 2026-08-14
Recent updates
bilibili_extract tool; full transcript + danmaku; auto-ASR; frame vision via local backends; two-pass workflow; robust rate-limit handling.
FAQ
- Does it bundle models or binaries?
- No — no third-party binaries or models ship with the plugin; the open-source projects and services it invokes are listed in THIRD_PARTY_NOTICES.md.
- Can it run fully offline?
- Yes — switch ASR to local sherpa-onnx (SenseVoice) or whisper.cpp, and vision to a local Ollama/llama.cpp backend; one failing source never breaks the rest.
- How does the agent see frames?
- It reads the transcript first, then requests frames with explicit timestamps; each frame is captioned with nearby subtitles and the sharpest frame near each target wins.
Alternatives
dhicoc/dsh-reverse-skill · FuzzySoul/dsh-free-vision · Anionex/dsh-vision-toolkit