• Primary: General users / AI enthusiasts (want to learn how to use AI)
• Secondary: Fellow product managers (will examine content from a professional perspective, so content must stand up to scrutiny)

1. First-person narrative
   • Write from the "I" perspective: "I recently discovered... when working on X" "My view is..."
   • This is a product manager's personal sharing, not an official document
2. Clear opinions with solid evidence
   • Dare to express positions: "I think A is more suitable for this scenario than B"
   • But must provide reasons and evidence, don't make empty claims
3. Practice-oriented
   • Talk less about "what it is", focus more on "how to use it", "what pitfalls I encountered", "what scenarios it's suitable for"
   • Use real cases and scenarios to illustrate points
   • Must include real usage scenarios: Don't pile up data or list parameters, instead demonstrate value with "what I actually used it for"
   • Show specific usage processes and results, including both successful and failed cases
4. Cover image is mandatory
   • Every article must have a theme cover image generated
   • Use Simplified Chinese for text on the image
   • Adopt a left-right split layout (text on the left, visual elements on the right)
   • Must clear the ALL_PROXY environment variable when calling the Gemini API (see Step 6 for details)
5. Content structure diagram is mandatory
   • Every article must have a content structure diagram/infographic generated
   • Place it at the beginning of the article (after the cover image)
   • Adopt a Graphic Recording style
   • Used to display the overall structure and core points of the article
6. Use plain text format for links
   • ❌ Wrong:

     [Official Website](https://example.com/)

   • ✅ Correct:

     Official Website: https://example.com/

• "Why did Cursor become popular? Teardown of its product design"
• "What makes Perplexity's search experience great?"
• "Notion AI vs Obsidian + AI, which is more suitable for you?"

• Must include real usage scenarios: What did I actually use it for? How was the effect?
• What problem does this product solve?
• What is the core function design logic? (Combined with usage experience)
• Who are the target users? Why can it impress them?
• What design ideas can be learned from it?
• What shortcomings do I think it has? (Based on actual usage)

• "Complete thinking for building a customer service robot with Dify"
• "How to use AI to automate competitor analysis?"
• "How I used RAG for knowledge base Q&A"

• What are the pain points of this scenario?
• Why choose this solution? (Compared with which alternatives)
• How to build/implement it specifically? (Steps + screenshots)
• How is the effect? What pitfalls were encountered?
• Who is it suitable for?

• "My workflow optimization insights after using Claude Code for a week"
• "5 tips to make Dify workflows more stable"
• "These Cursor shortcuts are unknown to 90% of people"

• Real usage scenarios: What specific tasks did I use this tool for? (Don't just talk about functions, talk about "what I did")
• What efficiency-enhancing skills were discovered? (With specific examples)
• What pitfalls were encountered? How were they solved?
• Share specific configurations/prompts/workflows
• Compare before-and-after effects: How much difference in efficiency with and without using it?

• "How much technology does an AI product manager need to understand?"
• "How to design an Agent with product thinking?"
• "My understanding of AI product MVP methodology"

• What is my view/methodology?
• Where does this view come from? (Experiences, cases, thinking)
• How to implement it specifically?
• Are there any counterexamples or boundary conditions?

• "Agent is so popular, but I think 90% of scenarios don't need it"
• "Will the MCP protocol change the landscape of AI applications?"
• "From a product perspective, what are the differences between Claude and ChatGPT?"

• What is this event/trend? (Quick popular science, but don't pile up data)
• What do I think? (Clear opinion): Based on my observations and thinking, not repeating others' views
• Why do I think this way? (Evidence and reasoning): Support with my actual experiences or observed cases
• What does it mean for general users/product managers? (Provide specific actionable suggestions)
• Important reminder: Don't write it as a news report or data dump, write "my opinion"

WebSearch

• "{Product name} official website", "{Product name} feature introduction"
• "{Product name} review", "{Product name} user reviews"
• "{Product name} vs {competitor}"

• "{Scenario} AI solutions"
• "{Tool name} {Scenario} tutorial"
• "{Scenario} best practices"

• "{Tool name} skills", "{Tool name} advanced usage"
• "{Tool name} workflow"

• "{Topic} product manager"
• "{Topic} methodology"
• Related cases and data

• "{Topic} latest news"
• "{Topic} industry analysis"
• Various viewpoints and discussions

WebFetch

1. Official documents/official blogs
2. Product-oriented media like Product Hunt, Sspai
3. Community discussions on Zhihu, Jike
4. Technical blogs (supplement technical details)

• Core functions and design philosophy of the product
• User feedback and usage scenarios
• Data and cases
• Different viewpoints and controversies

• First-person: "I" "we" "you"
• Colloquial but not casual: Like chatting with friends, but organized
• Short sentences mainly: No more than 25 words
• Use more specific cases, fewer abstract descriptions

• Dare to make judgments: "I think A is more suitable than B"
• But provide reasons: "Because..." "From my experience..."
• Acknowledge limitations: "Of course, this is just my view" "May not be applicable in XX scenario"

1. Proxy setting issue (most common)
   • Problem: Google Genai SDK does not support the

     socks5h://

     proxy protocol
   • Error message:

     Unknown scheme for proxy URL URL('socks5h://...')

   • Solution: Clear the

     ALL_PROXY

     environment variable before the command
   bash

   # ❌ Wrong: Will cause error
   python scripts/generate_image.py --prompt "..." --api gemini --output cover.png
   
   # ✅ Correct: Clear ALL_PROXY
   ALL_PROXY="" all_proxy="" python scripts/generate_image.py --prompt "..." --api gemini --output cover.png

2. Image format issue
   • Gemini API returns JPEG format (not PNG)
   • The script will automatically handle base64 decoding and binary saving correctly
   • The output file extension can be

     .png

     , but the actual content is JPEG
3. API key configuration
   • Ensure the

     GEMINI_API_KEY

     or

     GOOGLE_API_KEY

     environment variable is set
   • Missing key will cause error: "Please set the environment variable GEMINI_API_KEY or GOOGLE_API_KEY"
4. Prompt length limit
   • Gemini API has a limit on prompt length
   • If the prompt is too long, the API may return an error
   • It is recommended to keep the prompt within 2000 characters

cd /root/.claude/skills/wechat-product-manager-writer

• Has a clear viewpoint or core value
• Uses first-person narrative
• Has real usage scenarios (Mandatory! Not data piles or parameter lists)
• Shows specific usage processes and results (including success and failure)
• Viewpoints are well-founded and stand up to scrutiny
• Provides actual help to readers (can learn something or get inspiration)

• Language is easy to understand, no jargon stacking
• Short sentences mainly, easy to read
• Has own attitude and style
• Clear structure, reasonable subtitles

• Cover image has been generated
• Content structure diagram has been generated
• Links use plain text format
• Word count is between 1500-3000 words
• No redundant chapters like "References" "Further Reading"

Recently when working on the XX project, I encountered a problem: ...

Many people ask me: How much technology does an AI product manager need to understand? Let's talk about my views today.

I've always thought that Agent is overhyped in most scenarios.

I found a tool two days ago, after using it for a week, I want to share my usage insights.

To summarize: ... Hope this helps you.

This is just my current thinking, what do you think? Welcome to discuss in the comment section.

If you have similar scenarios, you might try this solution. Leave a message if you have questions.

1. Write in your own words, have personal style
2. Dare to express opinions, but provide reasons
3. Use more specific cases and scenarios
4. Acknowledge your own limitations and shortcomings
5. Provide readers with actual actionable suggestions

1. Don't write it as an official document or press release
2. Don't pile up feature lists and data (cold data like "6 million cost" "2 months development" has no value)
3. Don't talk about viewpoints without case support
4. Don't use too many professional terms (explain if used)
5. Don't write about things you don't understand
6. Most importantly: Don't write product analysis without real usage scenarios

1. Marketing gimmick cases: Such as "$1000/month startup project", "get rich overnight" etc.
2. Exaggerated comparison cases: Such as "Work that takes a team a year, AI finishes it in an hour"——even if sourced, this kind of comparison makes readers feel fake
3. Unverifiable cases: If you can't find the original link/source, better not use it
4. Second-hand retold cases: Media retelling often exaggerates and distorts, prioritize original sources

1. First-hand developer sharing: Such as "Boris Cherny said he generated 259 PRs with Claude Code"——has specific numbers, said by the person involved
2. Official data: Data from changelogs, official blogs, product documents
3. Verifiable tweets/posts: Can provide links
4. Own real experiences: This is the most credible

• Better use plain descriptions than exaggerated cases
• If a case sounds "too good to be true", readers will probably think the same
• When citing others' cases, ask yourself: Does this have an original link? Dare I put the link out?
• Better describe the function itself than use a suspicious "shocking" case

• Don't repeat the same viewpoint or technical point in different chapters of the article
• If a concept has been explained earlier, don't expand on it later
• After writing, check: Is there any passage that says something similar elsewhere?

• Not an encyclopedia, no need to cover everything
• Not an official document, must have personal perspective
• Not a marketing soft article, must have real value