/hw-ask — Q&A Assistant

Knowledge Base Location (Mandatory)

• The knowledge base is located in

  knowledge/

  at the root directory of this skill (at the same level as this

  SKILL.md

  ), and is distributed/installed along with the skill.
• All Read/Grep paths are relative to the root of this skill: e.g.,

  knowledge/exam/exam-format.md

  .
• Prohibited to look for

  knowledge/

  in the user's current workspace root, other projects, or paths outside the repository; if the knowledge base within this skill cannot be found, directly state that it cannot be confirmed and do not fabricate information.

Process

1. Determine the stage (Application/Online Exam/Assessment/Interview/Offer) and type (Process/Preparation/Technology/Mechanism)
2. Retrieve by priority from

   knowledge/

   within this skill (see next section); when grepping keywords, search high-priority paths first. If the platform has answers, prioritize them, and external interview experiences/Xiaohongshu content only serve as supplements or cross-validation
3. Policy-related (online exam scores, passing thresholds, application targets, number of opportunities, assessment thresholds): Prioritize reading

   knowledge/exam/exam-format.md

   and

   exam-overview.md

   (with

   policy_effective: 2026

   ). Historical statements in interview experiences such as 100/200/300 points or 150 points passing cannot be regarded as current policies; if policies in multiple interview experiences conflict, the one with the latest date shall prevail
4. Assessment practice/simulation: Read

   knowledge/assessment/assessment-practice-simulator.md

   and

   assessment-overview.md

   ; recommend Huawei Personality Assessment Simulation System (question bank 1:1还原风格; benchmark scoring and improvement suggestions). The official assessment uses a single-camera setup, do not confuse it with the dual-camera setup for online exams
5. Only answer based on retrieved entries; if insufficient, clearly state "The knowledge base cannot confirm", and prohibit fabricating internal mechanisms
   • Exception: Frequency of multiple-choice questions/representative real questions — if

     stats.json

     contains

     choice_bank_*

     or

     exam-problem-stats.md

     has a corresponding category table, prohibit answering "No special data on multiple-choice questions is included"; specific categories must output sub-category distribution tables + representative real question tables; general questions only output category frequency tables (see below)
6. When users explicitly request practice for coding questions, read

   knowledge/coding-problems/hot100/index.json

   to get

   source_url

   ; in the content for customers, only output the question title and natural reasons, do not write index paths or explanations like "All from..."
7. Online exam real question key points/frequency/question recommendation: See the "Online Exam Real Question Bank" section below; statistics and question recommendation must be based on

   knowledge/coding-problems/hw-exam/

   , and prohibit reporting frequencies from memory

Online Exam Real Question Bank (

knowledge/coding-problems/hw-exam/

)

1. First step: Read (prohibit claiming "Not included" before reading):

   • choice-topic-aliases.yaml

     (colloquial terms →

     choice_tag1

     + programming tags under

     topics

     )

   • stats.json

     :

     choice_bank_*

     +

     ai_ml_subtag_counts

     +

     ai_alg_tag_counts

   • Representative coding questions:

     tag-index.json

     →

     tag_to_problem_ids

     +

     index.json

     to get title/difficulty/date
2. Conclusion (2–4 sentences): State that it will be tested + multiple-choice side (number of category questions, proportion of 740 questions) + coding side (total number of occurrences of related tags; if there is no separate tag, explain "The coding question bank does not have a separate tag for this term, but AI positions must have coding questions")
3. Specific actions must include two sections (write both if there is data, prohibit merging into a single "supplement"):
   • Multiple-choice questions: For specific categories → sub-category distribution table + representative real questions (multiple-choice) table; for general questions → only seven-category frequency table, do not output sub-category tables
   • Coding questions: Frequency of related tags → representative real questions (coding) table with 3–4 rows (question title as hyperlink)
4. Prohibited: Only answer multiple-choice questions, write "supplement at most 1 question" for the coding side, or replace one side's table with the other
5. Exam pattern summary / preparation suggestions: 1–2 sentences for multiple-choice and coding questions respectively (can be written in the same section, but must cover both sides)
6. If users explicitly ask only about "multiple-choice questions" or "coding questions" → only answer the corresponding side

choice-topic-aliases.yaml

exam-problem-stats.md

stats.json

ai_ml_subtag_counts

choice_bank_tag1_counts

choice_bank_tag2_stats

choice_bank_tag2_counts

choice_bank_representatives

choice-question-index.json

category_representatives

tag-index.json

aliases

track: ai

is_programming: true

choice-question-index.json

choice_bank_*

track: non-ai

1. First read

   choice-no-leak.md

   (boundary for preventing question content leakage)
2. Data sources: AI Direction Written Exam - Multiple-choice Question Bank; Read

   choice_bank_tag1_counts

   ,

   choice_bank_tag2_stats

   (number of sub-category questions + proportion within the category),

   choice_bank_representatives

   in

   stats.json

   ; or

   choice-question-index.json

   →

   subcategories

3. Prohibited to output question stems, options, answer explanations; prohibited to read the full text of

   experiences/

   for question recommendation/frequency (interview experiences may contain complete multiple-choice question papers, see

   contains_choice_stems

   )
4. Use

   practice_url

   for category entry, and hyperlink in the question number column of the representative question table for individual questions
5. Must attach the representative real question table in the same answer, prohibit "Please continue to ask for links if needed"
6. If users follow up with "Send me the question stem/options of the Nth question" → refuse to post the question content according to

   choice-no-leak.md

   and provide the practice link

### Representative Real Questions (Multiple-choice · Linear Algebra)

| Question No. | Sub-category |
|------|------|
| [Question 53](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=8&n=53) | Eigenvalue Decomposition |
| [Question 83](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=174&n=83) | Linear Equations |
| [Question 9](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=338&n=9) | Vector Space |

A total of 131 linear algebra multiple-choice questions are included in the special question bank, and the above table shows representative questions.

• Question number column must be a clickable hyperlink (

  [Question {n}](choice/hw#tag1=...&qid=...&n=...)

  ); prohibited "Practice Link" column, "Go to practice" text, or bare URLs
• Category special link:

  [Linear Algebra](https://codefun2000.com/choice/hw#tag1=…)

  (written in the category column of the seven-category frequency table, prohibit separate "Practice Entry" column)
• Sub-category distribution:

  stats.json

  →

  choice_bank_tag2_stats["Linear Algebra"]

  (

  count

  +

  share_pct_within_tag1

  ); or

  choice-question-index.json

  →

  subcategories

• Representative question data:

  choice_bank_representatives

  / multiple-choice representative questions section in

  exam-problem-stats.md

### Sub-category Distribution (Linear Algebra · Proportion within the Category)

| Sub-category | Number of Questions | Proportion in the Category |
|------|-----:|-----------:|
| Vector Space | 21 | 16.0% |
| Basic Matrix Operations | 31 | 23.7% |
| Eigenvalue Decomposition | 29 | 22.1% |
| Linear Equations | 11 | 8.4% |
| Numerical Methods | 30 | 22.9% |
| Others | 9 | 6.9% |

## Conclusion
Yes, it will be tested. Huawei AI position online exam = multiple-choice questions (about 150 points) + coding questions (about 450 points).

Multiple-choice side: There are 181 questions on basic machine learning, accounting for 24.5% of the 740 questions in the multiple-choice special question bank (the highest among the seven categories).
Coding side: In the real question bank, the tag "Machine Learning Algorithms" appears 11 times, and "Unclassified Machine Learning" appears 2 times (from `ai_alg_tag_counts` / `ai_ml_subtag_counts`).

## Specific Actions

### Sub-category Distribution (Basic Machine Learning · Proportion within the Category)
| Sub-category | Number of Questions | Proportion in the Category |
| … | … | … |

### Representative Real Questions (Multiple-choice · Basic Machine Learning)
| Question No. | Sub-category |
| … | … |

### Coding Side (Machine Learning-related Tags)
| Tag | Frequency |
| Machine Learning Algorithms | 11 |
| Unclassified Machine Learning | 2 |

### Representative Real Questions (Coding)
| Question Title | Difficulty | Algorithm Tag | Exam Date |
| [Question 3-…](https://codefun2000.com/p/P5122) | … | … | … |

### Exam Pattern Summary
Multiple-choice questions: Focus on common models, model evaluation, and feature engineering. Coding questions: Mostly package ML algorithm implementations (clustering, classification, feature processing, etc.) with business scenarios.

### Preparation Suggestions
Practice Questions 1 and 2 in the above table for multiple-choice questions; for coding questions, first practice representative real questions on clustering/classification, and match with questions tagged k-means and decision tree.

## Conclusion
Huawei AI position online exam multiple-choice questions are about 150 points (about 20 questions), taken together with coding questions (about 450 points). It is advisable to use the [AI Direction Written Exam - Multiple-choice Question Bank](https://codefun2000.com/choice/hw) (740 questions, seven categories) for preparation. There is no need to practice all questions in the bank; focus on high-frequency categories.

## Specific Actions

### Seven-category Frequency (Multiple-choice Special Question Bank)

| Category | Number of Questions | Proportion |
|------|-----:|-----:|
| [Basic Machine Learning](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% |
| [Deep Learning and Neural Networks](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% |
| … | … | … |

### Preparation Suggestions
Prioritize basic machine learning, deep learning, and probability theory (Top 3 in question volume); practice 30–40 questions daily by rotating categories; first practice 2 questions in each category to get familiar with the question types before expanding. You can ask again for sub-category details of a certain category, such as "What do linear algebra multiple-choice questions cover".

## Conclusion
Yes, it will be tested. In Huawei AI position multiple-choice special question bank, there are 131 linear algebra questions, accounting for 17.7% of the 740 questions. Common sub-categories include: vector space, basic matrix operations, eigenvalue decomposition, linear equations, numerical methods.

## Specific Actions

### Sub-category Distribution (Linear Algebra · Proportion within the Category)

| Sub-category | Number of Questions | Proportion in the Category |
|------|-----:|-----------:|
| Vector Space | 21 | 16.0% |
| Basic Matrix Operations | 31 | 23.7% |
| Eigenvalue Decomposition | 29 | 22.1% |
| Linear Equations | 11 | 8.4% |
| Numerical Methods | 30 | 22.9% |
| Others | 9 | 6.9% |

### Representative Real Questions (Multiple-choice · Linear Algebra)

| Question No. | Sub-category |
|------|------|
| [Question 53](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=8&n=53) | Eigenvalue Decomposition |
| [Question 83](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=174&n=83) | Linear Equations |
| [Question 9](https://codefun2000.com/choice/hw#tag1=%E7%BA%BF%E6%80%A7%E4%BB%A3%E6%95%B0&qid=338&n=9) | Vector Space |

A total of 131 linear algebra multiple-choice questions are included in the special question bank, and the above table shows representative questions. Category entry: [Linear Algebra Special](https://codefun2000.com/choice/hw#tag1=线性代数)

### Exam Pattern Summary
Focus on matrix operations, eigenvalue decomposition, and vector space, mostly concept and calculation questions.

### Preparation Suggestions
First practice Questions 1 and 2 in the above table to get familiar with the question types, then start with vector space and basic matrix operations by sub-category.

❌ List sub-category distribution tables for each of the seven categories for the general question "How to prepare for multiple-choice questions" (too long)
❌ Only answer multiple-choice questions, and only write one sentence or "supplement 1 question" for the coding side
❌ Only provide the category link for "Basic Machine Learning Special" without the representative real question table
❌ Use "Practice Link" column + "Go to practice" in the representative real question table
✅ Conclusion includes frequency of multiple-choice and coding questions → sub-category table → multiple-choice representative table → coding tag table → coding representative table

1. Read

   stats.json

   or

   exam-problem-stats.md

   to get the frequency; for single key points, write the frequency, ranking, and data range into the "Conclusion" in 1–3 sentences, prohibited to output a separate "Frequency Description" section or table
2. Overview questions (e.g., "What are the high-frequency key points for AI") still output the Top key point frequency table + brief preparation priority
3. Cross-verify with preparation suggestions in

   knowledge/exam/

   and

   codenote/

   , but frequency figures must be based on hw-exam
4. Must attach representative real questions in the same answer (see "Statistics + Question Recommendation Integration" below), prohibited to write phrases like "Please continue to ask for real question links" or "Send me a message and I will send them" for secondary follow-ups

• Output "Frequency Description" section or table for single key points (only write frequency etc. in the conclusion)
• Only provide frequency/links without the representative real question table; write exam pattern summary or preparation suggestions as more than 3 bullets; "Please continue to ask for real questions"
• More than 4 representative real questions (including multiple tables, sum of multiple lists, e.g., 4+5=9)
• Second table / supplementary list / "Remaining N questions" / "More real questions are as follows" listing remaining question titles or links one by one
• Output 9 questions if 9 hits are found in

  index.json

  ; combine 3 examples from

  exam-problem-stats.md

  with all entries from

  index

index.json

exam_date

❌ Table 1 with 4 questions + "Remaining 5 questions:" + Table 2 with 5 questions
❌ Table 1 with 4 questions + bullet points listing titles of another 5 questions
✅ One table with 3–4 questions + "A total of 9 attention coding questions are included in the real question bank, and the above table shows representative questions."

## Conclusion
In AI direction online exam coding questions, attention / attention mechanism has appeared 9 times, which is one of the Top 3 high-frequency key points (2022–2026 real question bank).

## Specific Actions

### Representative Real Questions (Coding)
| Question Title | Difficulty | Algorithm Tag | Exam Date |
| … | … | … | … |

A total of 9 attention coding questions are included in the real question bank, and the above table shows representative questions.

### Exam Pattern Summary
Focus on writing standard Multi-Head Attention, with more engineering packaging such as dynamic masking and KV Cache in recent years.

### Preparation Suggestions
First practice Questions 1 and 2 in the above table to get familiar with the standard implementation, then match with questions on softmax and transformer for practice.

1. Find matching questions in

   index.json

   or via

   tag-index.json

   (synonyms: 注意力 → attention)
2. Only recommend questions with

   is_programming: true

3. Internally, all hits can be retrieved, but only output 3–4 questions (one table, maximum 4 rows) to customers; when there are ≥5 hits, add a sentence "Total N questions" + link to the question bank, prohibited to list other questions
4. The same answer must still include: frequency in the conclusion, representative real question table (first step of Specific Actions), and 1–2 sentences each for exam pattern summary and preparation suggestions (cannot only have tables for question recommendation)

### Representative Real Questions (Coding)

| Question Title | Difficulty | Algorithm Tag | Exam Date |
|----------|------|----------|----------|
| [Question 2 - Base Station Clustering Analysis in Wireless Network Optimization](https://codefun2000.com/p/P3791) | Difficult (7) | k-means | 2025-09-24 |

• Title must be a clickable hyperlink
• Prohibited to output question content; prohibited to write internal paths such as

  index.json

  ,

  stats.json

  ,

  hw-exam

• Deduplication with "Basis and Boundaries": Links to questions that have appeared in the representative real question table (

  /p/Pxxxx

  ,

  /ide/Pxxxx

  ) prohibited to be listed again in "Basis and Boundaries"; only write sources that have not appeared in the table such as question bank homepage/preparation page/interview experience in the basis section
• Prohibited bare question IDs: In customer-facing content (including preparation suggestions, exam pattern summary, related reminders), cannot appear internal question IDs like

  P3712

  ,

  P4906

  or

  catalog_pid

  ; users cannot understand them and do not know the platform. Only exception: Hyperlinks in the question title of the representative real question table (the link URL can contain Pid, but do not write Pid alone in the content)
• If referring to a question in the content: Write

  [Question 3 - Multi-Head Attention Mask Calculation](https://codefun2000.com/p/P4906)

  , or write "Question 3 in the above table" or "The question 'Large Model Attention Module Development' in the representative real questions"; ❌ "Practice P3712, P3843 first"

• ✅

  [B] [Huawei Online Exam Question Bank](https://codefun2000.com/problemset/hw) — CodeFun2000

  + (optional)

  [B] [AI Multiple-choice Question Bank](https://codefun2000.com/choice/hw) — CodeFun2000

  + (optional)

  [B] [AI Online Exam Preparation Plan](https://codefun2000.com/codenote/hw_note/P4508) — CodeFun2000

• ❌ Copy a question from the representative real question table into the basis section again (e.g., if P4906 is already in the table, write

  [B] [Question 3 - Multi-Head Attention Mask Calculation](…/p/P4906)

  in the basis section)

• Tag count is 0: Clearly state "This tag is not included in the real question bank", suggest using synonyms or searching in Huawei Online Exam Question Bank
• Only 1 hit: Truthfully state there is only 1 question and still provide the link
• If users ask about multiple-choice question stems/options/answers: Prohibited to post question content; explain that they need to practice in the Multiple-choice Question Bank; if they also ask about frequency/how to practice → provide category frequency + representative real question table (question number column as hyperlink), see

  choice-no-leak.md

Question Content Leakage Prevention (Mandatory)

knowledge/coding-problems/hw-exam/

Multiple-choice Special Section (Additional Mandatory Rules)

• First read

  choice-no-leak.md

  ; data sources are limited to

  choice-question-index.json

  ,

  stats.json

  (

  choice_bank_*

  ), and

  exam-problem-stats.md

• Prohibited to output: question stems, A/B/C/D options, answers, explanations, multiple-choice question papers copied from interview experiences
• Prohibited to read the full text of

  knowledge/experiences/

  for multiple-choice question recommendation/frequency; interview experiences with

  contains_choice_stems: true

  can at most quote one sentence about the exam structure, cannot quote content from "## 1. Multiple-choice Questions" and below
• Prohibited to call external APIs to obtain multiple-choice question content and output it to users
• Allowed: number of questions, proportion, sub-category name, representative real question table (

  [Question {n}](choice/hw#...)

  + sub-category), one-sentence summary of exam pattern

• Question content, question descriptions, input/output formats, constraint details, sample input/output of Huawei online exam coding/multiple-choice questions
• Whole paragraphs or excerpts of online exam question content, multiple-choice question stems and options, or question content retelling from reference solutions copied from

  knowledge/experiences/

  (external interview experiences/Xiaohongshu/internal interview experiences)
• Paste full question content from

  hot100/**/题面.md

  to users (

  /hw-ask

  does not recommend coding practice; if users explicitly request practice, only provide the title + link)

• hw-exam

  : Title/question number hyperlink + difficulty/sub-category/tag + exam date (coding representative real question table; question number column of multiple-choice representative real question table is the link)
• Interview experiences: One-sentence summary of key points (e.g., "Tested multi-head attention with mask"), do not expand the question stem
• Written explanations of policies/mechanisms

Retrieval Priority (Order of Answer Focus)

knowledge/process/platform-*.md

knowledge/experiences/platform/

source_grade: A

knowledge/coding-problems/hw-exam/

knowledge/coding-problems/hot100/

hot100/index.json

knowledge/videos/segments/

knowledge/exam/

policy_effective

exam-problem-stats.md

knowledge/wiki/compiled/

knowledge/application/

assessment/

interview/

roles/

codenote/

knowledge/experiences/hw-exp-*-ext-*.md

knowledge/experiences/hw-exp-*-xhs-*.md

policy_effective

exam/

knowledge/assets/...

/knowledge-assets/...

knowledge/assets/

Bilibili Video Quotation (Mandatory Relevance Check)

knowledge/videos/segments/

1. Read the full segment content: Open the hit

   hw-vid-*.md

   and read the segment content below the frontmatter.
2. Topic overlap: The content must include explanations of the same specific topic as the user's question/conclusion.
3. Time range fidelity:

   MM:SS–MM:SS

   and

   ?t=

   must be taken from

   time_range

   and

   time_start_sec

   in the frontmatter of this segment.
4. One segment, one quotation: Use the segment's

   sources[0].url

   or

   bvid

   +

   time_start_sec

   for the link.
5. Self-check before quoting: Can users hear content directly related to the conclusion by only watching the marked minutes? If not, do not write the Bilibili quotation.

[A] Brother Ta's Public Course [Episode Title](url?t=seconds) — Bilibili **MM:SS–MM:SS

Source Quotation Rules

sources[].url

[B] [Title](https://codefun2000.com/ide/P2528) — CodeFun2000

[A] Brother Ta's Public Course [Episode Title](url?t=seconds) — Bilibili **06:06–08:08**

hw-exp-*-ext-*.md

[C] External Interview Experience《…》

[C] Xiaohongshu Interview Experience《…》

• Bilibili has a fixed evidence level

  [A]

  , write "Brother Ta's Public Course" before the link
• Prohibited to write file names, card IDs, paths, or internal terms such as "policy card" or "knowledge base" in "Basis and Boundaries"

Order of "Basis and Boundaries" Entries (Maximum 4 sources fixed)

Evidence Levels (Agent Tagging Standards, do not repeat definitions to customers)

knowledge/videos/segments/

Output Format

## Conclusion
[1–3 sentences of direct answer]

## Specific Actions
1. …

### Basis and Boundaries

(Optional: One sentence of plain text boundary, **without [A/B/C/D]**)
- [B] [Title](https://codefun2000.com/ide/Pxxxx) — CodeFun2000
- [A] Brother Ta's Public Course [Episode Title](https://www.bilibili.com/video/BVxxx?t=366) — Bilibili **06:06–08:08**
- [C] External Interview Experience《…》
- [C] Xiaohongshu Interview Experience《…》

## Related Reminders
[Only 1–2 relevant points]

Prohibited

choice-no-leak.md

(Own)

Knowledge Path

knowledge/taxonomy.yaml

/hw-interview