ASU-style Real Technical Resume Generation Skill

Generate Chinese technical resumes that are "information-dense at first glance, and still correspond to real experiences item by item on second glance". Default for Xiaohongshu long images, technical homepages and internal displays; when users explicitly request a job-seeking version, reduce dramatic wording while retaining the same facts

Default Deliverables

```
resume-data.json
```
: Structured resume driven by real sources.
```
resume.html
```
: Self-contained webpage suitable for full-page browser screenshots.
```
resume.pdf
```
: 2–3 page A4 PDF.
```
xiaohongshu-copy.md
```
: Publishing copy based on real experiences.

Generation Principles

Retain Real Skeleton Directly

The following information authorized by users can be used as-is:

Name, school, degree, company, department, position and dates;
Real project names, technical solutions, papers/patents/shares;
Metrics clearly stated in documents with interpretable caliber;
First-hand statements from users about internal responsibilities.

Do not automatically change real names to "a certain company" or "a certain school", nor add "fictional" after each item. Only replace when users request anonymization.

Only Enhance Expression, Do Not Create Facts

Allowed:

Organize real technical processes into

Input -> Representation -> Planning -> Tools -> State -> Evaluation

links;

Concentrate real models, frameworks, training strategies and deployment components from the same project;

Use terms consistent with the project such as

Workflow / ReAct / Harness / Context Engineering / Benchmark / ModelOps

;

Summarize real evolution with macro mainlines like "from model nodes to Agent systems to long-term continuous operation";
Create humorous contrast through titles, parallelism, rhythm and density of technical terms.

Prohibited:

Add non-existent educational backgrounds, titles, offers, salaries, projects, open-source roles or business results;
Change "participated" to "led/Owner/core author";
Present planned work as implemented;
Rewrite original denominators or add different metrics together;
Use company, project stars or team achievements to imply personal completion of all results.

Write Role Strength According to Sources

Source states "independently responsible" → Prioritize writing
```
module-level Owner / end-to-end link Owner / from 0→1
```
, while limiting to specific modules;
Source states "mainly responsible" → Can write
```
strategy architecture Owner / direction leader
```
, with specific responsibility boundaries retained in the main text;
Source states "participated in promotion/co-construction" → Write as
```
system-level co-builder / core participant
```
, do not forge as global Owner;
Source states "plan/future direction" → Mark as
```
current direction / in planning
```
, cannot be changed to launched results.

Each key project must first undergo an "Ownerization split": find the smallest closed loop that the user truly independently completed or was mainly responsible for, such as

data governance link

strategy architecture

quantitative deployment

distribution diagnosis

. Owner and 0→1 can be used for this closed loop, but the scope of these terms cannot be expanded to the entire company platform.

School Title Enhancement

For educational experiences, must check school official websites, education authorities or formal university alliance pages to extract real but high-potential institutional labels:

Construction sequence:

Double First-Class / Former 211 / Jointly Built by Ministry and Municipality / Research University

;

Historical positioning:

First university in the country / Founding year

;

University alliances: Real membership relations such as
```
Group of Eight / C9 / Russell Group
```
;
Degree facts: Undergraduate, master's, joint training and degree-granting institutions must be separated, cannot write geographical location or cooperation projects as non-existent degrees.

The title adopts

School Name | Institutional Title 1 | Institutional Title 2 | Degree

, and write the source URL into

source_note

. Do not write Australian degrees as "US undergraduate", nor write jointly built by ministry and municipality as "directly under central ministries".

Visuals and Wording

Read

references/style-guide.md

before generation. Must have:

Blue serif section titles, light red/light blue/light green experience bars;
Company, department, date, direction tags on the same line;
Projects fixed use "Background / Metrics and Effects / My Responsibilities / Technical Keywords";
Each key project has at least 5–10 professional terms consistent with the source;
Projects that can be established from the source must include
```
Owner
```
,
```
0→1
```
and end-to-end closed loop with limited Scope; if not applicable, use
```
co-builder/core participant
```
instead;
Responsibility bullets prioritize using technical chains, failure classification, data closed loops or architecture evolution;
Complete sections such as education, internship/work, technical projects and precipitation, awards and skills;
Do not add watermarks, disclaimers or "fictional" marks.

Workflow

1. Complete Reading of Experience Sources

Read resumes, Feishu documents, screenshots or personal homepages provided by users. For long documents, first take the table of contents, then read completely by chapter. Record for each experience:

Factual identity: Organization, department, position, dates;
Role boundaries: Independently responsible, mainly responsible, participated or planned;
Project objectives, technical difficulties, solution links;
Metrics, baselines, results and time windows;
Model, framework, training, deployment, data and evaluation terms;
Shares, patents, awards and public materials.

For each school, additionally search official websites, education authorities and formal university alliance pages to form a three-column fact table of

Title -> Official Original Text -> URL

, then select 2–4 strongest and non-repetitive tags to include in the resume.

2. Establish Source Mapping

Each visible bullet uses:

json

{
  "text": "Build NL2SQL training and deployment link through structured short COT, real execution GRPO and AWQ quantization.",
  "verification": "source_grounded",
  "source_note": "Promotion Document / NL2SQL Chapter"
}

Do not add content not present in the document. If it is user口述 but cannot be publicly verified, use

verification: user_attested

3. Refine Technical Mainline

Prioritize selecting real evolution that can cover multiple experiences, such as:

Multimodal content understanding → Workflow model nodes → Search/general Agent → Context/Memory/Harness → Long-term Coding Agent

The mainline can be grand, but each node must have a corresponding project in the main text.

4. Generate Owner / 0→1 / Scope Narrative

Generate three levels of expression for each project:

```
Owner Scope
```
: The smallest closed loop that the user independently or mainly responsible for;
```
0→1 Action
```
: Establishing model, strategy, training, deployment or evaluation link from scratch; if only optimizing existing systems, rewrite as
```
X→Y Architecture Evolution
```
;
```
Large Scope
```
: List the real Data / Model / Training / Serving / Agent / Eval layers covered by this closed loop, do not count adjacent team results as personal results.

Example:

text

NL2SQL ModelOps Full Link Owner: From 0→1,打通 Executable SQL Cleaning -> Structured Short COT -> GRPO Execution Reward -> AWQ -> SGLang Serving, covering data, training, reinforcement learning, quantization and inference deployment.

5. Stack Professional Terms

Extract terms from sources, do not add randomly:

Models and representation: CLIP, BERT, LanguageBind, VideoLLaVA;
Training: LoRA, Full SFT, DPO, GRPO, Short COT;
Inference deployment: AWQ, vLLM, SGLang, Prefill, Decoding;
Agent: ReAct, Tool Use, Session, Memory, Context Engine, Checkpoint;
Evaluation: Suite, Case, Grader, Transcript, Outcome, Artifact.

Terms in the same bullet must have real logical relationships, cannot just list nouns.

6. Generation and Verification

Generate JSON according to

references/resume-schema.md

, then run:

bash

python3 scripts/render_resume.py \
  --input /absolute/path/resume-data.json \
  --html /absolute/path/resume.html \
  --pdf /absolute/path/resume.pdf

python3 scripts/validate_resume.py \
  --data /absolute/path/resume-data.json \
  --html /absolute/path/resume.html \
  --pdf /absolute/path/resume.pdf

After verification, convert each page of PDF to PNG, check for missing Chinese characters, cropping, unbalanced pagination and residual watermarks. For the webpage version, perform a full-page browser screenshot check.

Attached Materials

```
references/style-guide.md
```
: Layout and high-density technical wording rules.
```
references/resume-schema.md
```
: Source-driven JSON format.
```
assets/resume_template.html
```
: Self-contained Chinese template.
```
scripts/render_resume.py
```
: HTML/PDF generator.
```
scripts/validate_resume.py
```
: Source marking, metrics and file integrity check.
```
evals/evals.json
```
: Chinese test tasks.

asu-resume-skill

NPX Install

Tags

SKILL.md Content (Chinese)