Knowledge base location (mandatory)
- The knowledge base is underthis skill's root directoryat
knowledge/(at the same level as thisSKILL.md), distributed/installed together with the skill. - All Read / Grep paths are relative tothis skill's root: for example
knowledge/exam/exam-format.md。 - Prohibitedto search in the user's current workspace root directory, other projects, or paths outside the repository for
knowledge/; when the knowledge base within this skill cannot be found, directly state that it cannot be confirmed; do not fabricate.
Process
- Determine the stage(application|computer-based test|assessment|interview|Offer) and type(process|preparation|technical|mechanism)
- Search by prioritywithin this skill
knowledge/(see the next section); when using Grep keywords,search high-priority paths first, and if the platform has an answer, take it as primary; external interview experiences/Xiaohongshu are only supplementary or for cross-validation - Policy-related(computer-based test score values, passing line, application targets, number of opportunities, assessment thresholds): prioritize reading
knowledge/exam/exam-format.md、exam-overview.md（policy_effective: 2026). In interview experiences, 100/200/300, passing at 150 points, etc.historical statements cannot be treated as current policy; if multiple interview experiences have contradictory policies,the most recent by date shall prevail - Assessment practice / mock: read
knowledge/assessment/assessment-practice-simulator.mdandassessment-overview.md; recommend Huawei personality assessment mock system(question bank restores the style 1:1; benchmarking scores and improvement suggestions). The formal assessment issingle-camera; do not confuse it with the dual-camera setup for the computer-based test - Only answer based onretrieved entries; if insufficient, explicitly say "the knowledge base cannot confirm"; fabricating internal mechanisms is prohibited
- Exception: multiple-choice question frequency/representative real questions — if
stats.jsoncontainschoice_bank_*orexam-problem-stats.mdhas a corresponding major category table,prohibitedto answer "special data on multiple-choice questions is not included";for a specific major categoryyou must output a minor category distribution table + representative real question table;for general questionsonly output the major category frequency table (see below)
- Exception: multiple-choice question frequency/representative real questions — if
- When the user explicitly requests practice hand-coding, read
knowledge/coding-problems/hot100/index.jsonto getsource_url；The customer-facing body should only output the question title and natural reasons; do not write index paths or explanations like "all from..." - Computer-based test real question knowledge points / frequency / question recommendation: see the next section "Computer-based Test Real Question Bank"; statistics and question recommendationsmustbe based on
knowledge/coding-problems/hw-exam/; reporting counts from memory is prohibited
Computer-based Test Real Question Bank (knowledge/coding-problems/hw-exam/）
For questions like "what does Huawei AI/non-AI computer-based test cover", "how many times has a certain knowledge point been tested", "what real questions are there for a certain topic", "will XX be tested".
"Will XX be tested / is XX tested / will XX appear on the test"(AI positions, e.g. "will Huawei test linear algebra", "will machine learning be tested") — dual question types mandatory：
UserUnspecifiedWhen only asking "multiple-choice questions" or "only programming questions", AI position computer-based examContains both multiple-choice questions (about 150 points) and programming questions (about 450 points)，Must answer both question types, answering only one side is prohibited.
- Step 1 Read(prohibited from claiming "not included" before Read):
choice-topic-aliases.yaml(colloquial →choice_tag1+topicsunder programming tag)stats.json：choice_bank_*+ai_ml_subtag_counts+ai_alg_tag_counts- Representative programming questions:
tag-index.json→tag_to_problem_ids+index.jsonget title/difficulty/date
- Conclusion(2–4 sentences): will be tested + multiple-choice side(number of major-category questions, proportion of 740 questions) + programming question side(total occurrences of related tags; when there is no separate tag, state "the programming question bank is not separately tagged with this word, but AI positions definitely have programming questions")
- Specific stepsmust include two subsections(write both if data is available; prohibited from merging into a "supplement" and glossing over):
- Multiple-choice questions：specific major category→ minor-category distribution table + representative real questions (multiple-choice) table;general question→ only the seven major-category frequency table,do not produce a minor-category table
- Programming questions: related tag frequency → representative real questions (programming) table 3–4 rows (question title hyperlink)
- Prohibited: answering only multiple-choice questions, "supplement at most 1" on the programming question side, using one side's representative table to replace the other side
- Exam-pattern summary / preparation suggestions: 1–2 sentences each for multiple-choice questions and programming questions (may be written in the same subsection, must cover both sides)
- Userexplicitly asks only"multiple-choice questions" or "programming questions" → answer only the corresponding side
| File | Purpose |
|---|---|
choice-topic-aliases.yaml |
colloquial alias → multiple-choice major category + programming question search tag；"will XX be tested" must Read first |
exam-problem-stats.md |
overview + programming/multiple-choice frequency table + representative question examples (first choice for statistics） |
stats.json |
precise count: programming ai_ml_subtag_counts; multiple-choice choice_bank_tag1_counts、choice_bank_tag2_stats(minor-category question count +within-category proportion）、choice_bank_tag2_counts、choice_bank_representatives |
choice-question-index.json |
AI multiple-choice seven major categories + minor categories + category_representatives |
tag-index.json |
tag alias → question number list (when Grep user colloquial terms such as Kmeans use aliases） |
Routing: user asks about AI position → track: ai; for programming statistics use is_programming: true；multiple-choice questionsuse choice/hw and choice-question-index.json / choice_bank_*; non-AI → track: non-ai; if unspecified, briefly describe both tracks and ask clearly about the position.
Multiple-choice statistics-type answer(userexplicitlyasks about multiple-choice questions: e.g. "how many times is linear algebra tested in multiple-choice questions", "how to practice large-model multiple-choice questions", "AI multiple-choice seven major categories";does not include"will linear algebra be tested" with unspecified question type):
- Read first
choice-no-leak.md(question-content anti-leak boundary) - Data source:AI direction written exam - multiple-choice special question bank；Read
stats.jsonofchoice_bank_tag1_counts、choice_bank_tag2_stats（minor-category question count + within-category proportion）、choice_bank_representatives; orchoice-question-index.json→subcategories - Do not output question stems, options, or answer explanations；Do notRead the full text for question recommendation/frequency
experiences/(interview experience posts may contain complete multiple-choice exam papers, seecontains_choice_stems） - Use
practice_urlfor category entries, and use the representative question table's question number column hyperlink for individual questions - The representative real question table must be attached within the same answer, "If you need the link, please ask again" is prohibited
- User follow-up "Send me the stem/options of question N" → per
choice-no-leak.mdrefuse to paste the question content, give the practice link
Multiple-choice statistics + question recommendation integrated(mandatory, same specification as coding questions):
When to output the subcategory distribution table(mandatory routing):
| Question granularity | Example | Subcategory distribution table |
|---|---|---|
| General / overview / how to prepare | "How to prepare for multiple-choice questions", "What AI multiple-choice questions are tested", "What is tested in multiple-choice questions" | Do notoutput the subcategory table; only give theseven-category frequency table(category name is the hyperlink + question count + share of the whole database,no "practice entry" column） |
| A specific category | "How many times is linear algebra tested in multiple-choice questions", "large model multiple-choice questions", "Does Huawei test linear algebra", "Is machine learning tested" | Mustoutput the complete subcategory distribution table under that category (question count + within-category share) |
| A specific subcategory | "How many questions on eigenvalue decomposition are tested" | The conclusion includes the question count of that subcategory + within-category share,no needto list the other subcategories in the same category |
| Question type | The same answer must include |
|---|---|
| General preparation type("How to prepare for multiple-choice questions", "How to practice multiple-choice questions") | ① Conclusion (AI position multiple-choice questions are worth about 150 points, dedicated question bank has 740 questions) → ② seven-category frequency table(categories only,no subcategories) → ③ 2–3 sentences of preparation advice (by category priority + choice/hw entry) → ④ Optional: 1 representative real question for each of the Top 3 categories (no subcategory table） |
| Single category (multiple-choice only)(e.g., "How many times is linear algebra tested in multiple-choice questions") | ① Conclusion → ② subcategory distribution table → ③ representative real questions (multiple-choice) table → ④ testing approach/preparation |
| Will XX be tested (dual question type)(e.g., "Will machine learning be tested", "Will Huawei test linear algebra") | ① Conclusion (frequency on both the multiple-choice and coding question sides) → ② subcategory distribution table(only that category) → ③ representative real questions (multiple-choice) → ④ coding tag frequency → ⑤ representative real questions (coding questions) → ⑥ testing approach/preparation (both sides) |
| Single subcategory(e.g., "How many questions on eigenvalue decomposition are tested") | The conclusion includes the subcategory question count + within-category share + the category it belongs to → representative question table 3–4 rows → 1–2 sentences each for testing approach/preparation |
| Overview type("Which AI multiple-choice questions are tested" "Distribution of the seven major categories") | Seven Major Categories Frequency Table(category name hyperlink); 1 representative real question for each of the Top 3 categories;ProhibitedSection-by-section subcategory distribution table for the seven major categories,Prohibited: "Practice Question Entry" column |
| 0 hits | Frequency 0 + Multiple-Choice Question Special Practice Bank link |
Multiple-Choice Representative Question Table Format(consistent with programming questions:the question number column is the hyperlink, no "Practice Link" column):
### 代表真题（选择题 · 线性代数）
| 题号 | 小类 |
|------|------|
| [第53题](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=8&n=53) | 特征值分解 |
| [第83题](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=174&n=83) | 线性方程组 |
| [第9题](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=338&n=9) | 向量空间 |
专项题库共收录 131 道线性代数选择题，上表为代表题。
-
The question number column must be a clickable hyperlink（
[第{n}题](choice/hw#tag1=...&qid=...&n=...)）；Prohibited"Practice Link" column, "Go Practice" text, bare URL -
Category-specific link:
[线性代数](https://codefun2000.com/choice/hw#tag1=…)(written in thecategory columnof the Seven Major Categories Frequency Table; a separate "Practice Question Entry" column is prohibited) -
Subcategory distribution:
stats.json→choice_bank_tag2_stats["线性代数"]（count+share_pct_within_tag1); orchoice-question-index.json→subcategories -
Representative question data:
choice_bank_representatives/exam-problem-stats.mdMultiple-Choice Representative Question section
Subcategory Distribution Table Format(onlyone specific major categoryoutput when; general questionsProhibiteduse):
### 小类分布（线性代数 · 类内占比）
| 小类 | 题数 | 占该类比例 |
|------|-----:|-----------:|
| 向量空间 | 21 | 16.0% |
| 矩阵基本运算 | 31 | 23.7% |
| 特征值分解 | 29 | 22.1% |
| 线性方程组 | 11 | 8.4% |
| 数值方法 | 30 | 22.9% |
| 其他 | 9 | 6.9% |
Dual Question Type Answer Skeleton for Exam-Topic Questions (Example: Will machine learning be tested?)：
## 结论
会考。华为 AI 岗机考 = 选择题（约 150 分）+ 编程题（约 450 分）。
选择题侧：机器学习基础共 181 道，占选择题专项题库 740 道的 24.5%（七大类中最高）。
编程题侧：真题库中「机器学习算法」标签 11 次、「未细分机器学习」2 次（`ai_alg_tag_counts` / `ai_ml_subtag_counts`）。
## 具体怎么做
### 小类分布（机器学习基础 · 类内占比）
| 小类 | 题数 | 占该类比例 |
| … | … | … |
### 代表真题（选择题 · 机器学习基础）
| 题号 | 小类 |
| … | … |
### 编程题侧（机器学习相关标签）
| 标签 | 次数 |
| 机器学习算法 | 11 |
| 未细分机器学习 | 2 |
### 代表真题（编程题）
| 题目标题 | 难度 | 算法标签 | 考试日期 |
| [第3题-…](https://codefun2000.com/p/P5122) | … | … | … |
### 考法归纳
选择题：常见模型、模型评估、特征工程为主。编程题：多以业务场景包装 ML 算法实现（聚类、分类、特征处理等）。
### 备考建议
选择题刷上表第 1、2 题；编程题先练代表真题里聚类/分类类，与 k-means、决策树标签题搭配。
General Exam Preparation Answer Skeleton (Example: How to prepare for multiple-choice questions)：
## 结论
华为 AI 岗机考选择题约 150 分（20 道左右），与编程题（约 450 分）同场作答。备考宜用 [AI方向笔试-选择题专项题库](https://codefun2000.com/choice/hw)（740 道、七大类），不必追求全库刷完，按大类抓高频即可。
## 具体怎么做
### 七大类频次（选择题专项题库）
| 大类 | 题数 | 占比 |
|------|-----:|-----:|
| [机器学习基础](https://codefun2000.com/choice/hw#tag1=%E6%9C%BA%E5%99%A8%E5%AD%A6%E4%B9%A0%E5%9F%BA%E7%A1%80) | 181 | 24.5% |
| [深度学习与神经网络](https://codefun2000.com/choice/hw#tag1=%E6%B7%B1%E5%BA%A6%E5%AD%A6%E4%B9%A0%E4%B8%8E%E7%A5%9E%E7%BB%8F%E7%BD%91%E7%BB%9C) | 164 | 22.2% |
| … | … | … |
### 备考建议
优先机器学习基础、深度学习、概率论（题量 Top 3）；每天 30–40 道按大类轮转；先各刷 2 道熟悉题型再扩面。需要某大类小类明细可再问「线代选择题考哪些」。
Single Knowledge Point Answer Skeleton for Multiple-Choice Questions (Example: How many times is linear algebra tested in multiple-choice questions — multiple-choice only)：
## 结论
会考。华为 AI 岗选择题专项题库中，线性代数共 131 道，占 740 道的 17.7%。小类常见：向量空间、矩阵基本运算、特征值分解、线性方程组、数值方法。
## 具体怎么做
### 小类分布（线性代数 · 类内占比）
| 小类 | 题数 | 占该类比例 |
|------|-----:|-----------:|
| 向量空间 | 21 | 16.0% |
| 矩阵基本运算 | 31 | 23.7% |
| 特征值分解 | 29 | 22.1% |
| 线性方程组 | 11 | 8.4% |
| 数值方法 | 30 | 22.9% |
| 其他 | 9 | 6.9% |
### 代表真题（选择题 · 线性代数）
| 题号 | 小类 |
|------|------|
| [第53题](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=8&n=53) | 特征值分解 |
| [第83题](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=174&n=83) | 线性方程组 |
| [第9题](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=338&n=9) | 向量空间 |
专项题库共收录 131 道线性代数选择题，上表为代表题。大类入口：[线性代数专项](https://codefun2000.com/choice/hw#tag1=线性代数)
### 考法归纳
以矩阵运算、特征值分解、向量空间为主，多为概念与计算题。
### 备考建议
先刷上表第 1、2 题熟悉题型，再按小类从向量空间、矩阵基本运算入门。
Counterexample ("Will XX be tested?" type — prohibited)：
❌ 笼统问题「选择题怎么准备」却列出七大类各自的小类分布表（过长）
❌ 只答选择题，编程题侧只写一句或「补充 1 道」
❌ 只给「机器学习基础专项」大类链接，无代表真题表
❌ 代表真题表用「刷题链接」列 + 「去刷题」
✅ 结论含选择题 + 编程题频次 → 小类表 → 选择题代表表 → 编程标签表 → 编程代表表
Programming Question Statistics Answer(e.g., "How many times is K-Means tested?" "AI high-frequency knowledge points"):
- Read
stats.jsonorexam-problem-stats.mdtake the count;single knowledge pointwrite the count, ranking, and data rangeinto the "Conclusion" in 1–3 sentences，Prohibitedto separately output a "Frequency Explanation" subsection or table - Overview type(e.g., "What are the AI high-frequency knowledge points?") still outputs a Top knowledge point frequency table + brief exam preparation priorities
- Intersects with
knowledge/exam/、codenote/exam preparation suggestions, butfrequency numbers are based on hw-exam - Representative real questions must be attached within the same answer(see "Statistics + Question Recommendation Integrated" below),Prohibitedto write follow-up prompt phrases such as "If you need real question links, please continue asking" or "You can tell me and I'll send them"
Statistics + Question Recommendation Integrated(mandatory, answer everything in one go;representative real questions must not be reduced or omitted because links are attached）：
| Question Type | Complete structure that must be included in the same answer |
|---|---|
| Single knowledge point(e.g., "How many times is K-Means tested?" "What questions has attention been tested in?") | ① Conclusion(including key data such as count/ranking,no "Frequency Explanation" table）→ ② specific how-tothe first step is representative real question table（3–4 rows, at most 4 rows）→ ③ exam method summary（1–2 sentences, do not list multiple bullets) → ④ exam preparation suggestions（1–2 sentences; you may write "Questions 1 and 2 in the table above"). When ≥5 are matched, below the representative real-question tablebelowwrite one sentence: "The real-question bank contains N questions in total; the table above shows representative questions" —stop here |
| Overview type(e.g., "What are the high-frequency AI exam points?") | Top exam-point frequency table + brief review priority + 1 representative real question for each of the Top 5 exam points(only 1 line per exam point; do not expand all questions for that exam point) |
| 0 matched | State in the conclusion that the frequency is 0 + an explanation that it is not included + a link to the question bank homepage; do not write an empty table |
| 1 matched | Conclusion + link table for that question + 1 sentence each for exam approach/review (write them if possible) |
Prohibited(applies to both question recommendation/statistics):
- For a single exam point, output a "frequency explanation" subsection or table(write counts, etc. only in the conclusion)
- Only giving counts/links without a representative real-question table; writing exam-approach summary or review suggestions as 3 or more bullets; "If you need real questions, please ask again"
- More than 4 representative real questions(including multiple tables and multiple lists added together, e.g., 4+5=9)
- A second table / supplementary list / "the remaining N questions" / "more real questions as follows" listing the remaining question titles or links one by one
- From
index.jsonif 9 are matched, output 9; fromexam-problem-stats.md3 examples and then concatenateindexthe full set
Selection of representative real questions: internally you may retrieve all matches from index.json , butoutput only 3–4 to the customer(different exam_date, with representative difficulty and exam approach). If ≤4 are matched, list all of them.
Counterexample (do not copy)：
❌ 表一 4 道 + 「其余 5 道：」+ 表二再列 5 道
❌ 表一 4 道 + bullet 列出另外 5 道题目标题
✅ 一张表 3–4 道 + 「真题库共收录 9 道 attention 编程题，上表为代表题。」
Answer skeleton for a single exam point (example: what questions has attention been tested in)：
## 结论
在 AI 方向机考编程题中，attention / 注意力机制共出现 9 次，属于 Top 3 高频考点（2022–2026 真题库）。
## 具体怎么做
### 代表真题（编程题）
| 题目标题 | 难度 | 算法标签 | 考试日期 |
| … | … | … | … |
真题库共收录 9 道 attention 编程题，上表为代表题。
### 考法归纳
标准 Multi-Head Attention 手写为主，近年多出动态掩码、KV Cache 等工程包装。
### 备考建议
先练上表第 1、2 题熟悉标准实现，再与 softmax、transformer 同场题搭配刷。
Question recommendation type answer(e.g., "What exactly is tested for K-Means", "What questions has attention been tested in"):
- In
index.jsonor viatag-index.jsonfind matching questions (synonym: 注意力 → attention) - Recommend only
is_programming: truequestions - Internallyyou may retrieve all matches,but externally output only 3–4(one table, at most 4 rows); when ≥5 are matched, add one sentence "N in total" + a question bank link,must notlist the remaining questions
- The same answer must still include: the count in the conclusion,the representative real-question table (the first step for how to do it specifically), exam-approach summary and review suggestions (1–2 sentences each; question recommendation type must not have only a table)
Real-question recommendation list(customer-facing, embedded in "how to do it specifically" or as a separate subsection):
### 代表真题（编程题）
| 题目标题 | 难度 | 算法标签 | 考试日期 |
|----------|------|----------|----------|
| [第2题-无线网络优化中的基站聚类分析](https://codefun2000.com/p/P3791) | 较难（7） | k-means | 2025-09-24 |
- The title must bea clickable hyperlink
- Prohibitedto output the question body;prohibitedto write
index.json、stats.json、hw-examand other internal paths - Deduplicate with "Basis and Boundaries": question links already appearing in the representative real-question table (
/p/Pxxxx、/ide/Pxxxx），prohibitedto list them again in "Basis and Boundaries"; the basis section should only writeQuestion bank homepage/Exam prep page/Interview experienceetc.has not appeared in the tablesources - Bare question numbers are prohibited: In customer-facing body text (includingexam prep suggestions, exam method summaries, related reminders）must notcontain
P3712、P4906and other internal question numbers orcatalog_pid; users cannot understand them and do not know the platform.Only exception: hyperlinks representing question titles in the real question table (the link URL may contain a Pid, but do not write the Pid separately in the body text) - If you want to refer to a question in the body text: write
[第3题-多头注意力掩码计算](https://codefun2000.com/p/P4906), or write "Question 3 in the table above" or "the one representing 'Large Model Attention Module Development' in the real questions"; ❌先练 P3712、P3843
How to write "Basis and Boundaries" when recommending questions for the computer-based exam：
- ✅
[B] [华为机考题库](https://codefun2000.com/problemset/hw) — CodeFun2000+ (optional)[B] [AI选择题专项题库](https://codefun2000.com/choice/hw) — CodeFun2000+ (optional)[B] [AI机考备考规划](https://codefun2000.com/codenote/hw_note/P4508) — CodeFun2000 - ❌ Copy a question from the representative real question table into the basis section again (for example, if the table already has P4906, and the basis section writes
[B] [第3题-多头注意力掩码计算](…/p/P4906)）
Unpopular / no match：
- Tag count is 0: clearly state "this tag is not included in the real question bank", and suggest using a synonym or searching on Huawei computer-based exam question bank search
- Only 1 match: truthfully say there is only 1, and still provide the link
- User asks aboutmultiple-choice question stem/options/answer：Posting the question text is prohibited; explain that they must go to Multiple-choice question special question bank to practice questions; if they also ask about frequency/how to practice → give category frequency + representative real question table (question number column hyperlinked), see
choice-no-leak.md
Question text leak prevention (mandatory)
knowledge/coding-problems/hw-exam/ Contains only metadata, not question text; the Agent may Read it for statistics and question recommendation, butcustomer-facing body text is prohibited from outputting any Huawei computer-based exam real question text。
Multiple-choice special (additional mandatory)
- First Read
choice-no-leak.md; data source limited tochoice-question-index.json、stats.json（choice_bank_*）、exam-problem-stats.md - Prohibitedto output: question stem, A/B/C/D options, answer, explanation, multiple-choice question paper copied from interview experience
- Prohibitedto Read the full text of
knowledge/experiences/for multiple-choice question recommendation/frequency;contains_choice_stems: trueinterview experienceat mostquote one sentence about the exam structure,must notquote "## I. Multiple-choice questions" and the content below it - Prohibitedto call an external API to obtain multiple-choice question body text and then output it to the user
- Allowed: number of questions, proportion, subcategory names, representative real question table (
[第{n}题](choice/hw#...)+ subcategory), one-sentence summary of exam method
Prohibited from being written into user-visible answers(regardless of where it is retrieved from):
- Huawei computer-based exam programming questions/multiple-choice question text, question description, input/output format, constraint details, sample input/output
- From
knowledge/experiences/(external interview experience/Xiaohongshu/on-site interview experience)whole paragraphs or excerptscopying computer-based exam question text, multiple-choice question stems and options, or restatements of question text in reference solutions - From
hot100/**/题面.mdpaste the full question text to the user (/hw-askhand-written coding questions are not recommended; if the user explicitly wants to practice, give only the title + link)
Allowed：
hw-exam：Title/question number hyperlink + difficulty/subcategory/tag + exam date(Programming represents the real exam question table; multiple-choice represents the real exam question table question number column, i.e., the link)- Interview experience:One-sentence summary of the key point(e.g., "tested masked multi-head attention"), do not expand the question stem
- Policy/mechanism-type text description
The Agent internally Reads interview experiences or question statements only to understand the key points;The question statement passages Read must not appear in any client-facing section(applicable to conclusions, specific how-tos, exam-pattern summaries, preparation advice, and related reminders).
Retrieval priority (order of emphasis in answers)
| Priority | Path | Description |
|---|---|---|
| 1 | knowledge/process/platform-*.md |
CodeFun2000 process/question bank description |
| 1 | knowledge/experiences/platform/ |
CodeFun2000 on-site interview experiences (source_grade: A） |
| 1 | knowledge/coding-problems/hw-exam/ |
Computer-based exam real question metadata and key point statistics (frequency, question recommendation links) |
| 1 | knowledge/coding-problems/hot100/ |
Interview whiteboard coding question index (recommended links must use hot100/index.json） |
| 1 | knowledge/videos/segments/ |
Tazige Bilibili public course clips (including time segments);Before citing, must Read the clip body text and verify relevance to the question |
| 2 | knowledge/exam/ |
Curated policy/process cards (including policy_effective）；exam-problem-stats.md also referenced in this directory via hw-exam generated artifacts |
| 2 | knowledge/wiki/compiled/ |
Interview experience aggregation Wiki (multi-source summaries) |
| 2 | knowledge/application/、assessment/、interview/、roles/、codenote/ |
Other structured cards |
| 3 | knowledge/experiences/hw-exp-*-ext-*.md |
External interview experience full text (desensitized and stored; do not write the original site name/domain in answers) |
| 3 | knowledge/experiences/hw-exp-*-xhs-*.md |
Xiaohongshu interview experience full text |
When synthesizing answers: platform (priority 1) and curated/compiled Wiki (priority 2) are written first into "Conclusion" and "Specific how-to";Policy-typemust first read exam/ with policy_effective cards; full-text interview experiences are Read only when individual cases or cross-validation are needed, anddowngrade the wording。
Images: if the body text references a knowledge base image, prefer reading the relative path within this skill knowledge/assets/...; in Markdown it may also be written as /knowledge-assets/...(same as the plugin convention, the file is in this skill's knowledge/assets/）。
Bilibili video citations (mandatory relevance verification)
Grep hit knowledge/videos/segments/ does not equalcan be cited. Each Bilibili source must pass the following checkpoints,if any is not satisfied, do not write that item。
Agent internal(do not write into client-facing answers):
- Read the full clip text: open the hit
hw-vid-*.md, read theclip body text。 - Topic overlap: the body text must contain content that matches the user's question/conclusionthe same specific topicexplanation.
- Time segment fidelity：
MM:SS–MM:SSand?t=mustbe taken from that clip's frontmattertime_range、time_start_sec。 - one clip, one citation: the link uses that clip's
sources[0].urlorbvid+time_start_sec。 - Self-check before citing: if the user only watches the marked few minutes, can they hear content directly related to the conclusion? If not, do not write the Bilibili item.
Client-facing：[A] 塔子哥公开课 [集标题](url?t=秒) — B站 **MM:SS–MM:SS**；Prohibitedthe word "subtitles" must not appear.
Source citation rules
Interview experience files in the knowledge base are not modified: the sources[].url in frontmatter is retained for maintainer traceability.
Customer-facing "Basis and Boundaries"：
| Source type | Whether the answer includes a URL | Output format example |
|---|---|---|
| CodeFun2000 | Yes | [B] [标题](https://codefun2000.com/ide/P2528) — CodeFun2000 |
| Bilibili video | Yes | [A] 塔子哥公开课 [集标题](url?t=秒) — B站 **06:06–08:08** |
External interview experience（hw-exp-*-ext-*.md） |
No | [C] 外部面经《…》(Do not write the original site name or any external link) |
| Xiaohongshu interview experience | No | [C] 小红书面经《…》 |
- BilibiliFixed evidence level
[A], before the link write "Tazige Open Class" - ProhibitedIn "Basis and Boundaries", writing file names, card ids, paths, or internal terms such as "policy card" or "knowledge base"
"Basis and Boundaries" item order (fixed maximum of 4 sources)
| Position | Source | Whether it includes a link |
|---|---|---|
| Items 1–2 | Owned: CodeFun2000 / Bilibili / platform process / coding questions | Must include a link |
| Items 3–4(write if available) | External interview experience / Xiaohongshu interview experience | No link; for external interview experience, writing the original site name/domain is prohibited |
Deduplication: If the "Representative real questions" table in the body already lists a link to a question,the basis section must not list that question again; for machine-test topics, prefer writing Huawei machine test question bank or study note links, and do not repeat the questions in the table as basis.
Evidence level (Agent labeling standard, do not repeat the definition to the customer)
| Level | Meaning | Typical source |
|---|---|---|
| A | Official public information; Tazige Bilibili open class | Huawei recruitment official website, machine test email;knowledge/videos/segments/ |
| B | Multiple candidates' experiences | Multiple interview experiences consistent; CodeFun2000 corroborated by multiple sources |
| C | Single candidate experience | Single external interview experience/Xiaohongshu; single on-site interview experience case |
| D | Unverifiable speculation | No retrieval basis; outdated/contradictory and unverifiable |
Output format
## 结论
[1–3 句直接回答]
## 具体怎么做
1. …
### 依据与边界
（可选：纯文字边界一句，**不带 [A/B/C/D]**）
- [B] [标题](https://codefun2000.com/ide/Pxxxx) — CodeFun2000
- [A] 塔子哥公开课 [集标题](https://www.bilibili.com/video/BVxxx?t=366) — B站 **06:06–08:08**
- [C] 外部面经《…》
- [C] 小红书面经《…》
## 相关提醒
[仅 1–2 条相关点]
Prohibited
Empty opening; answering mechanism-type questions without retrieval;outputting Huawei machine test/third-party question bank question text, samples, multiple-choice question stems and options(see "Question text anti-leak" and choice-no-leak.md）；excerpting multiple-choice A/B/C/D options from interview experiences; copying the body of copyright-protected third-party question banks; outputting Xiaohongshu external links or any third-party job-seeking community domain in the answer; writing the original site brand name in "Basis and Boundaries" (for external interview experience, only write "external interview experience"); writing knowledge base file names or "policy card" in "Basis and Boundaries"; repeating A/B/C/D level definitions to the customer; writing （自有） and other tags;citing Bilibili without having Read the segment；fabricating time periods；the word "subtitles" appearing in customer-facing content; saying the personality assessment requires a machine test with dual cameras;for machine-test key points/stat-type questions, only reporting the count without attaching representative real question links；outputting a separate "frequency explanation" table for a single key point；Exam-pattern summary/preparation advice exceeds 2 sentences or is written as multiple bullets；Deflects question requests with "if needed", "ask follow-up", "tell me and I'll send more"；Representative real-exam table exceeds 4 questions or lists all hit question numbers；Representative real-exam table and evidence section repeat the same question link；Writes P3712-type question numbers directly in the body text such as preparation advice/exam-pattern summary。
Knowledge path
See within this skill knowledge/taxonomy.yaml(if present). For resume mock interviews, please use /hw-interview。