One-Stop Competitor Blogger Monitoring + Video Parsing

🎯 Pain Point: The store manager spends 2 hours daily browsing competitor bloggers, with no accumulated data and no clarity on which videos are worth learning from; secondary creation requires switching tools to download videos and parse scripts, resulting in a fragmented workflow.

Core Capabilities:

1. Monitor N bloggers → Algorithm scoring → Feishu card report of TOP 3
2. One-click script parsing within the same skill — Select viral videos → yt-dlp downloads videos → faster-whisper transcribes → DOUBAO multimodal parsing of 5 dimensions (shots / lines / visuals / sound / rhythm) → Output executable secondary creation md

II. Video Parsing Capability (Merged from original video-script-parser)

scripts/parse-video-script.py

# Single video parsing
python scripts/parse-video-script.py --url "https://www.douyin.com/video/xxx"

# Save locally only without writing to Feishu Wiki
python scripts/parse-video-script.py --url "..." --no-wiki

• 1. Shot Script (second-level granularity)
• 1. Copy & Lines (including transition points)
• 1. Visual Design (composition / color scheme / props)
• 1. Sound Design (narration rhythm / BGM selection)
• 1. Rhythm Analysis (climax position)

• yt-dlp

  to download videos from Douyin / Kuaishou / WeChat Channels

• faster-whisper

  local ASR (Alternative: DOUBAO ASR API, used when HF network is blocked)

• DOUBAO doubao-1-5-vision-pro-32k-250115

  multimodal visual frame parsing
• Anti-hotlinking: Solve CDN 403 with requests Referer + Cookie header

III. Typical Workflow (Seamless Monitoring → Parsing)

Automatically run monitor-bloggers.py at 9:00 daily
  ↓
Feishu card "Today's Top 3 Viral Videos"
  ↓
Store manager clicks the [Parse Script] button for one of the videos
  ↓
parse-video-script.py runs automatically
  ↓
Complete parsed md is received in Feishu Wiki after 30 seconds
  ↓
Zhu Jianhao creates new videos based on the md

I. 5 Algorithms + 4 Screening Rules (Replicate douyin-monitor AI Blogger Monitoring)

Algorithms (Viral Score 0-10)

4 Screening Rules (Meet any → Selected)

1. AI Score ≥7 (Subjective scoring by DOUBAO/DeepSeek)
2. Collect Rate ≥10% (High retention value)
3. Comment-to-Like Ratio ≥8% (High discussion)
4. Like-to-Fan Ratio ≥40% (Viral potential)

II. Prerequisites

pip install httpx openai
lark-cli auth login --scope "bitable:app:readwrite im:message:send"

# Configure credentials
export LARK_FASHION_COCKPIT_BASE_TOKEN="..."
export TABLE_BLOGGER_VIDEOS="tblxxxxxxxx"  # Table 27
export DEEPSEEK_API_KEY="..."
export DOUYIN_COOKIE="..."  # Used for crawling Douyin data

III. Initialization (Create table + populate mock data)

python scripts/seed-mock-bloggers.py
# → Populate 10 mock women's fashion blogger data (including real-style narration scripts) in Feishu Table 27

IV. Run Monitoring

# Full workflow: Crawl data → Calculate scores → Write to table → Send card
python scripts/monitor-bloggers.py

# Custom parameters
python scripts/monitor-bloggers.py --top 5 --reply-chat-id oc_xxx

V. Typical Report Card

👀 Competitor Blogger Daily Monitoring

📊 Monitored Bloggers: 10 | Videos Crawled Today: 10
🎯 Selected (Worthy of Secondary Creation): 3 | Pending Evaluation: 7

🔥 Top 3 Viral Scores (Sorted by Algorithm)

[1] Cheng Cheng Jie_Outfit Tutorial · 850k Fans · 🎤 With Narration
📝 150cm Height Enhancement & Slimming Tips | 3 Correct Ways to Tuck Shirts into High-Waisted Pants
👍 425k  💬 8.5k  ⭐ 56k  🔄 12k
Like-to-Fan Ratio 50.00%  Collect Rate 13.18%  Viral Score 9/10
💡 Recommendation Angle: Pain Point Solving via Tutorial
Narration Script (First 120 words): Hi everyone, I'm Cheng Cheng Jie. Today I'll teach you three correct ways to tuck shirts into high-waisted pants...
[▶ Watch Video]

[2] Sandy Little Apple · ...
[3] Plus-Size Girl Zihan · ...

[📋 View Complete Table 27]

VI. Collaboration with Other Skills

• Video parsing capability has been merged into this skill (see "II. Video Parsing Capability" above), one-click call to

  parse-video-script.py

  within the same skill

• content-pipeline

  — Selected videos serve as secondary creation material sources

• competitor-monitor

  — Trend analysis linkage

VII. Cron Scheduled Run (Recommended)

# Auto-run at 8:30 AM (paired with morning-report at 9:00)
crontab -e
30 8 * * * cd /path/to/lark-fashion-cockpit && python skills/blogger-monitor/scripts/monitor-bloggers.py

VIII. References

• Monitoring main script:

  scripts/monitor-bloggers.py

• Mock data script:

  scripts/seed-mock-bloggers.py

• Card style sample:

  examples/sample-card.md