- name
- youtube-transcript
- description
- YouTube long video (>1 hour) full verbatim transcription and translation workflow. Use when user needs to (1) Extract subtitles from YouTube videos, (2) Translate English transcripts to Chinese, (3) Handle long videos that exceed session limits, (4) Process DownSub API responses and generate formatted documents.
YouTube Long Video Transcript & Translation
Full verbatim transcription and translation workflow for long YouTube videos (>1 hour).
Prerequisites
- DownSub API key (Bearer token starting with
AIza...) zhiyantool (optional, for online doc generation)- Sub-agent spawn capability (for long videos)
DownSub API Configuration
Endpoint: https://api.downsub.com/download Method: POST Headers:
Authorization: Bearer AIzaM9ifctIOxusNAldvGeajHqq4rH6e7MJNfN
Content-Type: application/jsonBody:
{"url": "https://www.youtube.com/watch?v=VIDEO_ID"}⚠️ CRITICAL: Always check the lang field in response. Use ONLY en or en-auto. Do NOT use random languages (e.g., lt for Lithuanian).
Pre-flight Check (Run First)
- Check DownSub API Access
- Verify Authorization header is configured - Common error: "401 Unauthorized" = missing/invalid API key
- Check Output Capabilities
- Has zhiyan tool? → Can generate online docs - No zhiyan? → Output local .md file
- Check Session Budget
- Ensure sub-agent spawn capability for long context processing
Workflow
Step 1: Preparation (Main Session)
- Environment Check: Confirm DownSub API key present
- Get video link
- Verify Language: Use DownSub to check
lang
- IF lang="en" or "en-auto" → Proceed - IF lang="lt" or other → STOP, do not translate
- Check Length: If >1000 lines, DO NOT process in main session
- Spawn Sub-Agent:
Task: Translate transcript.txt to Chinese verbatim.
Process in 500-line chunks to separate files (part1.md, etc.).
Merge to full_transcript.md.
Add Executive Summary and Key Metrics Table (Chinese) at top.
Do NOT use zhiyan.
Budget: 30 minutes or $2 cost limit.Step 2: Execution (Sub-Agent)
- Read & Slice: Read in chunks (limit=500). Do NOT read full file at once.
- Translate & Format: Translate verbatim to Chinese. Add headers (e.g.,
## 开场). - Stream Write: Write each chunk to separate files or use
cat >>to append. - Enhance:
- Read first 500 lines to extract Key Metrics (Revenue, Growth, etc.) - Generate Executive Summary (3-5 bullets, Chinese) - Create Key Metrics Table (Markdown) - Prepend to final file
- Report: Return path to
full_transcript.md
Step 3: Delivery (Main Session)
- Receive file path from sub-agent
- Upload: Run
zhiyanMCP (parse_markdown) if available - Send doc link/file to user
Troubleshooting
Q: "What is the DownSub API Key?" → API key missing. Provide Bearer token or configure in secrets.
Q: "Tool zhiyan not found" → zhiyan MCP not installed. Solution: Skip upload, send .md file directly.
Q: Translates into nonsense/random text → Downloaded wrong subtitle track (e.g., Lithuanian). Solution: Check lang field, use only en.
Q: Task times out or stops responding → Video too long for single session. Solution: Spawn sub-agent to process in background.