Step 1: Rapidly Analyze Account Content Style
Launch Browser Automation Environment
Ensure your local environment has Chrome DevTools Protocol (CDP) configured for headless browser operations. Run this command to initialize dependencies:
'''bash bash ~/.claude/skills/web-access/scripts/check-deps.sh '''
This script checks Puppeteer/Playwright setup, port availability, and network connectivity to ensure smooth subsequent operations.
Access Creator Dashboard & Extract Note Data
Open the Xiaohongshu Creator Center note management page via a local proxy:
'''bash curl -s "http://localhost:3456/new?url=https://creator.xiaohongshu.com/new/note-manager" '''
Use the '/eval' command to extract titles and links of all published notes directly from the DOM, avoiding inefficient manual clicking.
Analyze High-Engagement Content Patterns
Identify the top 5 most-liked posts. Access each via in-page links (using '/click')—never construct URLs manually, to avoid 404s or triggering anti-bot detection.
After reading each, immediately run '/close' to close the tab. Keep active tabs under 3 at any time, adhering to browser resource management best practices.
Extract & Archive Style Elements
Analyze and summarize the following dimensions in under 200 characters:
- Opening pattern: Does it start with a question, pain point, or personal story?
- Sentence structure: Short sentences? Parallelism? Rhetorical questions?
- Tone: Warm big-sister vibe? Authoritative mentor? Casual吐槽 (sarcastic commentary)?
- Visual elements: Frequent emoji use? Fixed cover text format?
- Content length: Typical word count range?
Save output to:
'D:/ClaudeCode/company_asset/电子产品/从0到1PM面试/小红书宣传文章/小橙子风格分析.md'
Step 2: Consolidate Reference Materials & Extract Core Insights
Load Internal Content Repository
Read all non-empty '.txt' files from the local directory:
'D:/ClaudeCode/company_asset/电子产品/从0到1PM面试/小红书宣传文章/自己写的版本/'
These materials were authored by another account, “Mingjia Johnny,” featuring a sharp, direct, logic-heavy style (“incisive adjudicator” type). While unsuitable for direct use under the “Senior Sister Orange” persona, they contain highly validated PM interview knowledge frameworks.
Extract & Structure Key Points
Perform bullet-point extraction on each document to create a list structured like:
'''text
- Common STAR method pitfalls in behavioral interviews: Most candidates only describe Situation and Task, skipping detailed Action and quantified Result.
- Three-step framework for product design questions: User segmentation → Priority of needs → MVP design.
- Answering “What’s your biggest weakness?”: Pick a genuine but improvable trait, then tie it to a growth trajectory. ... '''
Note: Extract only factual insights—not tone or phrasing—to prepare for later style adaptation.
Step 3: Data-Driven Topic Planning
Conduct Keyword Popularity Research
Using CDP, perform searches on Xiaohongshu for these 5 keywords:
- PM面试
- 产品经理求职
- 面试技巧
- 简历优化
- 行为面试
Execution flow:
- Open new tab: '/new?url=https://www.xiaohongshu.com'
- Enter keyword in search bar
- Scrape top 5 high-like results—title and like count
- Close tab with '/close' after each search
Record trending subtopics under each keyword (e.g., “How to write a resume that passes screening,” “Common PM interview questions,” “Timeline for international students’ job hunt”) to identify audience interests.
Finalize 10-Post Content Calendar
Generate topics using three inputs:
- Current Xiaohongshu platform trends
- "Senior Sister Orange" content style
- High-signal content from Mingjia’s materials
Final topic mix:
| Type | Count | Example Topic |
|------|-------|----------------|
| Interview-focused (book-promotion) | 4 | 7 Must-Know PM Interview Questions — I Nailed All With This Book |
| Resume-focused | 2 | If HR Says These 3 Phrases, Your Resume Is Already Dead |
| Career/Growth | 2 | From 10 Rejections to FAANG Offer: My Mindset Reset Strategy |
| Open-topic | 2 | Are PM Roles in North America Actually Cutting Jobs? |
Export plan to:
'D:/ClaudeCode/company_asset/电子产品/从0到1PM面试/小红书宣传文章/小橙子排期计划.md'
Step 4: Batch-Generate On-Brand, High-Quality Content
Automate Writing via Codex Engine
To prevent style drift due to long context windows, offload bulk generation to an external specialized engine:
'''bash codex exec -m gpt-5.4-mini --full-auto --skip-git-repo-check " Context files: 1