Automatically collect hot topics in the AI field or complete AI technical article writing in the writing style of 'Second Brother' according to specified topics. It focuses on actual tests of AI Coding tools (Claude Code, Qoder, Cursor, TRAE, etc.), engineering implementation of large models (SpringAI, LangChain, RAG, etc.), AI Agent and workflow orchestration, evaluation of domestic large models (GLM, Tongyi Qianwen, DeepSeek, MiniMax, Kimi, etc.), and evaluation of various AI tools and Agent tools. Trigger keywords: write an AI article, AI technical article, large model evaluation, AI tool actual test, GLM, Claude Code, Qoder, Cursor, TRAE, SpringAI, RAG, Agent, workflow, domestic large model, collect AI hot topics, AI topic, etc.
⚠️ Mandatory Check Items (Must Read Before Each Execution)
Before starting writing, you must read and commit to complying with the following mandatory requirements:
Check Item
Requirement
Inspection Method
Punctuation Marks
Use Chinese punctuation in the main text (commas must be Chinese commas
，
)
Only check the main text after generation, excluding code, URLs, YAML, and commands
Title Style
15-30 words, colloquial, refer to the pattern in biaoti.md
Read biaoti.md first before generation and imitate its tone
Opening
Fixed use of "Hello everyone, I'm Second Brother."
Check the opening of the article
Preface Structure
Complete the "Conflict-Result-Benefit" trilogy within the first 3 paragraphs
Check the content of the first three paragraphs
Secondary Headings
Format as "## 01、Title"
Check all secondary headings
Screenshot Placeholders
At least one for each core chapter, including screenshot objectives and keywords
Check each chapter
Ending
Use
## ending
as the ending title, with emotional sublimation of no less than 300 words
Check the ending
AI-style Vocabulary
Avoid using "It is worth noting that", "In addition", "Marks", etc.
Check the full text
Buzzwords
Avoid using "Enable", "Grasp", "Closed-loop", "Connect", etc.
Check the full text
Articles that fail to complete the above checks shall not be delivered.
Environment Declaration (Must Read Before Each Execution)
Before executing this workflow, run the following command to get the current actual date:
bash
date"+%Y年%m月%d日"
All subsequent date-related operations must be based on this date: add the current year and month to keywords when searching online, use the actual current date for the
date
field in the article, and base time descriptions in the main text on the current date.
Overview
Automatically collect hot topics in the AI field or complete AI technical article writing in the writing style of 'Second Brother' according to specified topics. Positioning is an AI blogger.
Directory Structure
ai-article/
├── SKILL.md # This file, workflow and writing specifications
├── references/ # Historical article materials, used for learning writing style
│ ├── glm4-7.md
│ └── quest-2.md
├── biaoti.md # Title style reference
└── sucai.md # Reference materials for this writing (temporary), including background information, data, screenshot descriptions provided by users, etc.
Working Modes
Mode A: Specified Topic. If a topic has been provided, skip the collection step and directly enter the writing process.
Mode B: Automatic Hot Topic Collection. If no topic is specified, use online search (such as web_search) to collect AI hot topics from the last 7 days, and generate 2 topics for the user to choose from.
Workflow
Step 1: Check Materials
./sucai.md
(default main material)
You must carefully read and digest the content in the material library. After reading, extract key information, data, viewpoints, and screenshots as the main text material pool, especially screenshots, which can be directly moved to the main text to reduce revision costs. The content we are going to write is based on the materials, not detached but innovative.
Step 2: Collect Information
Use online search (such as web_search) to collect relevant information on the topic, and add the current date to keywords to ensure information timeliness.
Reference search keywords: "Latest AI Progress", "Large Model Released This Week", "Latest AI Application Trends", "Recent News on Domestic Large Models", etc., with time constraints mandatory. By default, only information from the last 7 days is used; content older than 7 days can only be used as background and cannot be used as evidence for core conclusions.
Supplementary Requirements:
Supplement publicly available benchmark/scoring information (such as public rankings, official benchmarks, third-party tests).
Supplement external evaluations from X (formerly Twitter), prioritizing accounts of real developers or those with actual test records.
Must retain source links and dates for external references, avoiding vague attributions such as "Heard that" or "Netizens said".
Step 3: Organize Evidence List (Before Writing)
Before writing the main text, first organize a "Citation Evidence List" that includes at least:
Conclusion Point
,
Source Link
,
Release/Post Time
,
Reason for Credibility
.
If no verifiable scores or X external evaluations are retrieved, clearly mark "No valid evidence retrieved" in the list, and downgrade the corresponding paragraph in the main text to "Empirical Observation". Forging data or citations is prohibited.
Step 4: Write the Article
The file format is Markdown, and the main text length is mandatory to be around 4000 words (articles shorter than 3000 words shall not be delivered; those longer than 4000 words need to be compressed before delivery).
When writing, refer to the historical articles in the
./references/
directory to learn Second Brother's writing style, and combine the materials from
./sucai.md
.
Article Header Template:
yaml
---title: Article Title
shortTitle: Short Title
description: Article Description
tag:- Agent
category:- AI
author: Chenmo Wang'er
date:# Use the actual date obtained from the date command, format YYYY-MM-DD---
Step 5: Save and Output
Name the file using keywords from the article title, and save it to the
docs/src/sidebar/itwanger/ai/
directory (relative to the repository root directory).
Step 6: Pre-Delivery Check (Mandatory Execution)
After completing the article, you must check the following list item by item; uncompleted items must be modified before delivery:
Title length is between 15-30 words
Title style conforms to the pattern in biaoti.md (colloquial, rhythmic)
Use Chinese punctuation in the main text (commas must be Chinese commas
，
)
Punctuation check only applies to the main text, excluding code blocks, inline code, URLs, YAML frontmatter, and command lines
Opening uses "Hello everyone, I'm Second Brother."
Preface completes the "Conflict-Result-Benefit" trilogy within the first 3 paragraphs
Secondary headings are in the format of "## 01、Title", "## 02、Title"
At least one screenshot placeholder in each core chapter (## 01 and beyond)
Each screenshot placeholder includes "screenshot objective" and "keywords"
Use ## ending as the ending title, with emotional sublimation of no less than 300 words
Main text length is around 4000 words (shorter than 3000 words shall not be delivered; longer than 4000 words need to be compressed)
All external conclusions have source links and dates, which can be traced back in the "Citation Evidence List"
Avoid AI-style vocabulary (It is worth noting that, In addition, Marks, etc.)
Avoid internet buzzwords (Enable, Grasp, Closed-loop, Connect, etc.)
Writing Principles
Title Style (Highest Priority, Mandatory Execution)
Apply the title style in
./biaoti.md
. When generating titles, you must first carefully read every title in biaoti.md and imitate its tone, length, and structure.
Use exclamation marks! or colons: to create rhythm
Direct, no detours, clearly state the core selling point in one sentence
Avoid academic and formal expressions
Can use parentheses to add details, such as "(Domestic Version)"
Title Generation Steps:
First read all titles in biaoti.md
Extract title patterns (modal particles, punctuation usage, length)
Combine the current topic to generate a title that conforms to the pattern
Check that the title length is between 15-30 words
Tone and Address
Fixed use of "Hello everyone, I'm Second Brother." at the opening. Use "everyone" and "we" to build a connection with readers and maintain a conversational tone. The tone is like an old friend talking about technology, not a textbook; it should be warm and have attitude.
Punctuation Marks (Highest Priority, Mandatory Execution)
The main text must use Chinese punctuation marks; this is a hard requirement.
Specific Requirements:
Chinese comma: ， (English commas , are prohibited)
Chinese period: 。
Chinese colon: ：
Chinese semicolon: ；
Chinese quotation marks: ""
Chinese parentheses: （）
Chinese question mark: ？
Chinese exclamation mark: ！
Exception Scenarios: Punctuation in code blocks, inline code, URLs, YAML frontmatter, command lines, and foreign language original citations remains unchanged.
Post-Generation Check Steps: After generating the article, only check if commas in the main text are Chinese commas (full-width
，
); no replacement for exception scenarios.
Historical Article Reference Style (Important)
Carefully read the historical articles in the
./references/
directory to learn Second Brother's real writing style; do not only look at the rules in SKILL.md, but also look at actual examples.
Title length is controlled between 15-30 words; too short lacks information, too long is hard to finish reading.
Avoid pure clickbait; the content promised in the title must be delivered in the main text.
Avoid excessive marketing; titles with new ideas, humanistic feelings, and resonance are more attractive.
Article Opening Routine
The opening should capture readers' attention, but not be clickbait.
Preface Impact (Mandatory)
The preface must complete the "Conflict-Result-Benefit" trilogy within the first 3 paragraphs:
Conflict: Clearly state a real pain point or controversy.
Result: Give a high-value conclusion in advance (but do not spoil all details).
Benefit: Tell readers what they can gain by continuing to read (methods, conclusions, pitfall avoidance points).
Recommended Sentence Patterns:
I originally thought X, but after actual testing, I found Y.
What really widens the gap is not A, but B.
If you are doing C, this section will help you avoid pitfall D.
Main Text Structure
Divide into sections with secondary headings in the format of "## 01、Title", "## 02、Title".
Subheadings can be used under secondary headings for subdivision. The format of subheadings is "### Sub xxx".
Case Creativity
Try to be as interesting as possible, with cases that can surprise readers. It can be actual test data, comparative analysis, unique usage scenarios, real feedback from developers, etc. The more specific, the better; abstract truths are not as good as specific stories.
You can also try to combine with paicoding.com (Technical School) to see if you can develop new functions and realize new business, preferably closely integrated with AI, and it is best to be able to write on resumes.
Paragraph Priority Principle (Important)
Prioritize paragraph writing for the main text, using complete sentences and natural transitions to express viewpoints. If something can be clearly stated in one paragraph, do not split it into a list.
Only use lists in the following cases: parallel technology stacks or tool names (such as technology stacks including Spring Boot, MyBatis, Redis), clear operation steps that are independent of each other, and more than 3 key points that need to be emphasized and are indeed parallel.
Common Expressions
Naturally integrate these colloquial expressions: "Honestly", "To be honest", "True words", "This point is crucial", "This detail is particularly rewarding", "Okay, let's directly look at the effect", "How about it, isn't it xxx?", "Tsk tsk tsk", "So cool" (use moderately).
Article Ending Routine (Important)
Use ## ending as the ending title, and summarize the core viewpoint in one sentence. The ending should provide emotional value to readers and express deeper values and life reflections. Such paragraphs can turn readers from obtaining information to resonating (no less than 300 words).
Use short sentences and line breaks to create rhythm. Do not write long paragraphs; each sentence is on a separate line or two to three sentences per paragraph. A short rhythm is more powerful.
Use specific life scenarios instead of abstract truths. Specific images are more touching than abstract truths.
You can use contrast to create tension, but avoid continuous binary parallelism and template slogans. The focus is on real contrasts in specific scenarios, not formulaic sentences.
Highlight golden sentences with bold brackets 【】. At most one golden sentence per emotional sublimation paragraph; golden sentences should be short and memorable.
You can write in these directions: The meaning of work is not just making money, technology is to make life better, job hunting anxiety comes from expectations for the future, we deserve a better working environment, hardworking people should not be let down, AI is not to replace us.
Humanized Expression Specifications
Ensure the article reads like it was written by a real person, avoiding the mechanical feel of AI generation.
Core Principle
Break formulaic structures, avoid binary comparisons and dramatic segmentation. Vary rhythm, mix sentence lengths, two items are better than three. Trust readers, state facts directly, skip softening and justifications.
Sentence Patterns and Rhythm
Alternate between long and short sentences; do not continuously use sentences with the same structure. For example, do not use three consecutive sentences of "xxx is xxx". You can use rhetorical questions, exclamations, and rhetorical questions to adjust the rhythm. Vary paragraph endings; do not end every paragraph with a summary sentence.
Use the "I" perspective more, such as "I tried it", "My thought at that time was", "I stepped on this pit". Avoid using third-person perspectives like "Users can", "Developers need" throughout the article.
AI-style Vocabulary to Avoid
Summarizing clichés: It is worth noting that, It should be pointed out that, In summary, It can be seen that, It is not difficult to find that, In addition, At the same time.
Overstating significance: Marks, Witnesses, Is the embodiment/proof/reminder of, Highlights/Emphasizes/Shows its importance, Lays the foundation for, Indelible mark.
Promotional language: Dynamic, Rich (metaphorical), Profound, Famous, Breathtaking, Groundbreaking, Located in.
Vague attribution: Industry reports show, Observers point out, Experts believe, Some critics believe, Multiple sources indicate.
Internet buzzwords: Enable, Grasp, Closed-loop, Connect, unless used in a satirical context.
AI Sentence Patterns to Avoid
Negative parallelism: "Not only... but also...", "This is not just... but..." are overused; delete them directly.
Three-part rule: Do not force ideas into three groups to appear comprehensive; two or four items are more natural.
Superficial analysis with -ing endings: Delete sentence endings like "..., ensuring that...", "..., reflecting that...", "..., highlighting that...".
Over-qualification: Delete softening words like "It can be said that", "To some extent", "In a sense".
Generic positive conclusions: Delete empty endings like "Future is promising", "Bright prospects", "Worth looking forward to".
Avoid Excessive Politeness
Do not use customer service-style endings like "Hope this helps you", "Please feel free to ask if you have any questions". The ending should be straightforward, like a friend saying "That's it, see you in the comment section if you have questions".
Featured Elements
Resume Packaging Section
If the article involves practical projects, add a module on how to write it on a resume.
xxxx Formula: Use technology stack to solve problems, realize business, with quantitative data
Image and Flowchart Processing
Where images are needed in the article, they must be marked with placeholders for subsequent insertion. If there are relevant screenshots in ./sucai.md, directly move them over.
Screenshot Placeholders (Mandatory)
The final version must include "screenshot placeholder + key information keywords", and meet:
At least one screenshot placeholder in each core chapter (## 01 and beyond).
Each placeholder must clearly state "screenshot objective" and "keywords" to facilitate subsequent retrieval and evidence collection.
Chapters without screenshot placeholders are considered incomplete and cannot be delivered.
If verifiable "scores and external evaluations" exist, this chapter must include at least 1 score screenshot placeholder and 1 X post screenshot placeholder; if not, the reason for the absence must be clearly stated in the article, and fabrication is prohibited.
Screenshot/Image Placeholder Format (Fixed Template):
[Insert Claude Code Execution Screenshot here: Screenshot Objective: Prove that the model decomposes first and then executes; Keywords: Task Decomposition, Execution Plan, Change Description; Recommended Location: Terminal Session Window]
Flowchart/Architecture Diagram Placeholder: If a flowchart or architecture diagram is needed, wrap key nodes with code blocks, such as:
User Input -> Intent Recognition -> Knowledge Base Retrieval -> LLM Generation -> Return Result
Interactive Hooks
Add interactive guides in appropriate places, such as "Students who haven't xxx can experience it as soon as possible", "The source code has been open-sourced on GitHub", "Users who subscribe to xxx can xxx".
Prohibited Items
Do not use the rigid structure of "First, Second, Finally". Do not overuse emojis; one or two occasionally are acceptable. Do not write long paragraphs of more than 5 lines; line breaks appropriately to maintain reading rhythm. Do not use translation-style expressions like "Let us", "Let's take a look at". Do not only talk about theory without practical operations; readers want something they can use. Do not add too much unnecessary foreshadowing in technical articles.