Douyin Transcript Capture to Feishu Bitable

Problem: Douyin captions are hard to archive

When engineers or operators receive a v.douyin.com link, they still need to listen, transcribe, segment, classify, and manually paste the result into Feishu. Video copy is not structured text, and the persuasion logic is spread across pacing, hooks, and spoken framing.

Workflow

The skill turns Douyin video transcription into a fixed pipeline:
- Link parsing: supports Douyin short links, www.douyin.com/video/, and /note/ image-text notes.
- Speech-to-text: uses the configured SILI_FLOW_API_KEY with SenseVoiceSmall to generate the raw transcript.
- Semantic segmentation: uses Qwen/Qwen2.5-7B-Instruct to split the transcript into structured paragraphs.
- Category recognition: outputs Level 1 > Level 2 > Level 3 based on Douyin's official third-level categories.
- Viral logic analysis: analyzes nine dimensions, including hook, pain point, selling point, pacing, and phrasing.
- Feishu write: posts to /bitable/v1/apps/{app_token}/tables/{table_id}/records, storing video title, link, category, raw transcript, segmented copy, and viral logic analysis.

Boundaries

• Requires SiliconFlow and Feishu credentials, and the Bitable schema must match the expected fields.
• The video link field must be an object like {text: "Open", link: "..."}; plain text may fail.
• The documentation hard-codes a macOS Apple Silicon FFmpeg path, so moving to another environment may break it.
• For /note/ content, the text may need to be fetched with info and then written to Feishu manually.