Resume Builder Cn

Resume Builder CN

Build truthful, targeted resumes from a reusable personal information library. Default to Chinese unless the user requests English or bilingual output.

Quick Start

Use the smallest workflow that satisfies the user's request.

If the user has no target role yet

Start with candidate intake and master profile construction.

Goal: build career-master-profile.md and identify missing information.

If the user has a target role, JD, job link, screenshot, company, team, or lab

Start with job intelligence and matching before writing.

Goal: produce job analysis, matching matrix, evidence selection, and a targeted resume.

If the user already has a resume

First inspect the existing resume, diagnose structure and evidence quality, then decide whether role research is needed.

If the user only wants critique

Do not generate a full resume package.

Output diagnosis, missing information, and prioritized revision suggestions.

If the user explicitly asks for deliverable files

Generate Markdown or HTML first. Generate PDF only when explicitly requested or when final delivery requires it.

Choose the Workflow

| User Input | Start From | Main Workflow | Default Output | |---|---|---|---| | "帮我看看简历有什么问题" | Resume inspection | Quick diagnosis | diagnosis + revision suggestions | | "帮我做一份简历" | Candidate intake | Master profile + resume draft | master profile + resume draft | | "我有这个岗位/JD/链接/截图" | Job intelligence | Targeted application package | job analysis + matching matrix + targeted resume + missing info | | "根据这个岗位优化我的简历" | Job intelligence, then evidence matching | Targeted rewrite | matching matrix + optimized resume | | "帮我生成 PDF/HTML 文件" | Existing resume or targeted draft | File generation and validation | HTML and optional PDF | | "做中英双语版本" | Language-specific drafting | Bilingual resume workflow | Chinese version + English version | | "帮我准备科研/学术简历" | Candidate classification | Research-oriented resume workflow | academic/research resume draft |

Where to Start

Start by reading only the rule files needed for the classified request:

• routing/classification:

[references/candidateclassification.md](references/candidateclassification.md);

• all cases: [references/intakeschema.md](references/intakeschema.md) and

[references/resumewritingrules.md](references/resumewritingrules.md);

• students/researchers:

[references/studentacademicrules.md](references/studentacademicrules.md);

• publications and patents:

[references/publicationsandpatentsrules.md](references/publicationsandpatentsrules.md);

• experienced candidates/career changers:

[references/experiencedcandidaterules.md](references/experiencedcandidaterules.md);

• targeted applications: [references/jobsourcerules.md](references/jobsourcerules.md),

[references/onlineroleresearch.md](references/onlineroleresearch.md), and the matching/evidence rules listed in the critical path;

• layout: [references/templatedesigns.md](references/templatedesigns.md);
• templates and examples: use assets/templates/ for required output structures

and examples/ only as format and quality references;

• content tuning, evidence trace table, bilingual version, photo prompt, and full

master profile JSON only when explicitly requested;

• final review:

[references/outputqualitychecklist.md](references/outputqualitychecklist.md).

Templates and Examples

Use templates in assets/templates/ to standardize output structure. Use examples in examples/ only as format and quality references. Never copy example facts, metrics, institutions, publications, awards, or experience content into a user's resume.

Templates define required sections and fields. Examples demonstrate acceptable reasoning depth, evidence traceability, bullet quality, layout density, and pass/fail quality standards.

When examples conflict with user evidence, job evidence, or non-negotiable rules, follow the user's evidence and the non-negotiable rules.

Mandatory Routing Step: Classify Before Writing

Before writing, rewriting, researching, or generating files, classify both the candidate and the resume scenario using [references/candidateclassification.md](references/candidateclassification.md).

Read [references/candidateclassification.md](references/candidateclassification.md) before selecting candidate rules or scenario-specific workflows.

This classification determines:

• which rules to read;
• whether online job research is required;
• whether the output should be a quick diagnosis, generic resume, targeted resume,

academic CV, or full deliverable package;

• which personal information should be included or omitted;
• whether ATS compatibility, visual layout, bilingual writing, or photo guidance is

relevant.

Classify the candidate:

• student;
• recent graduate;
• research-oriented candidate;
• experienced candidate;
• career changer.

Classify the scenario:

• Chinese resume;
• English resume;
• bilingual resume;
• academic/research CV;
• corporate job resume;
• internship resume.

Critical Path for Targeted Resumes: Job Intelligence and Matching

Run this path whenever the user provides:

• a job URL;
• JD text;
• JD screenshot;
• recruitment poster;
• company, team, lab, department, or research group name;
• target role title;
• request to optimize a resume for a specific position.

Do not draft the targeted resume before this path is completed or explicitly skipped due to lack of sources.

Read these files in order:

1. [references/jobsourcerules.md](references/jobsourcerules.md)
2. [references/jobrequirementextractionschema.md](references/jobrequirementextractionschema.md)
3. [references/candidateevidenceschema.md](references/candidateevidenceschema.md)
4. [references/matchingscoringrules.md](references/matchingscoringrules.md)
5. [references/experienceselectionrules.md](references/experienceselectionrules.md)
6. [references/traceabilityrules.md](references/traceabilityrules.md)

The module must preserve this chain:

job source -> requirement -> candidate evidence -> match score -> adoption decision -> resume bullet

Use stable IDs throughout the chain:

• job sources: JS-001, JS-002, ...
• job requirements: RQ-001, RQ-002, ...
• candidate evidence: EV-001, EV-002, ...
• resume bullets: BL-001, BL-002, ...

Each final bullet must include a hidden or separate trace record, for example:

BL-001 -> EV-003, EV-007 -> RQ-002, RQ-005 -> JS-001

Use the scripts as deterministic structuring aids. Their lexical scores are baselines, not substitutes for evidence-aware review.

If scripts are unavailable, perform the same schema transformation manually and clearly state that script validation was not run.

Decision Flow

Choose the smallest mode that satisfies the user request. Do not escalate to a larger deliverable set unless the user explicitly asks for those outputs.

Mode A: Quick Resume Diagnosis

Use when the user only asks for review, critique, or improvement suggestions.

Output: diagnosis, missing information, rewrite suggestions.

Mode B: Resume Rewrite

Use when the user provides an existing resume but no specific target role.

Output: revised resume content, optional Markdown/HTML.

Mode C: Targeted Application Package

Use when the user provides a JD, job link, company/team/lab name, or target role.

Run job intelligence, matching, evidence selection, and traceability.

Output: job-analysis-report.md, matching-matrix.md, targeted-resume.md or targeted-resume.html, and missing-info.md. Generate PDF only when requested.

Mode D: Full Deliverable Package

Use only when the user explicitly asks for files, HTML, PDF, editable templates, or complete package.

Output may include master profile, full master profile JSON, job report, matching matrix, resume HTML/PDF, trace table, content tuning sheet, bilingual version, and photo prompt according to the user's explicit request.

Use and Boundaries

Use for job/internship resumes, industry research resumes, concise academic CVs, resume optimization, bilingual conversion, and role-fit analysis. Photo guidance is auxiliary and should only be used when the user asks for resume/profile photo advice together with a resume task.

Do not use for government identity documents, visa/passport photos, fabricated credentials, deceptive employment histories, or full academic dossiers when the user needs a discipline-specific institutional format that has not been supplied.

Non-Negotiable Rules

• Never invent education, experience, metrics, publications, patents, awards, dates,

authorship, skills, or proficiency.

• Mark uncertain content as 待补充 or 需用户确认.
• Keep the complete master profile separate from each targeted resume.
• Trace every resume claim to personal evidence and every selection decision to

researched role evidence.

• Ask one coherent topic group at a time unless the user requests a form.
• Preserve accurate technical terminology and publication status.
• For publications and patents, preserve exact authorship role, author/inventor order,

status, DOI/patent number, journal/venue, partition/quartile, and verification source. Never upgrade unpublished, submitted, under-review, accepted, or patent-application status without evidence.

• Use action + task + method/tool + result/impact for experience bullets.
• Generate HTML as the primary editable, printable resume; generate Markdown when

requested or useful for review; generate PDF only when requested or clearly needed.

• When ATS compatibility is required, generate a plain Markdown or DOCX-friendly

version and do not rely on the HTML/PDF visual version as the only deliverable.

Workflow

1. Inspect and diagnose

Read supplied resumes, CVs, transcripts, notes, publications, portfolios, job links, and images before asking questions. Determine candidate type, scenario, output language, missing information, conflicts, and privacy preferences.

Do not ask for all missing information at once. First use available materials to produce:

1. current profile summary;
2. missing critical fields;
3. optional enhancement fields.

Only ask for information that changes the resume decision.

Run scripts/validate_intake.py when structured JSON is available.

2. Collect and normalize information

Follow references/intake_schema.md. Accept free-form conversation and summarize each topic for confirmation. Store complete candidate information only in the current working/output directory, never inside the skill package. Use assets/master_profile_template.md only as a blank template to copy or reference.

Use scripts/normalize_resume_json.py to normalize a JSON intake. Do not place personal data inside the skill directory.

3. Research the target online

For a specific application, follow references/job_source_rules.md and references/online_role_research.md even when the user supplies a JD.

Supported inputs include:

• official or third-party job URLs;
• team, laboratory, research-group, department, or project pages;
• JD screenshots, recruitment posters, official-account screenshots;
• copied JD text, HR descriptions, email text, or user summaries.

For screenshots, use OCR/vision to extract text and retain the screenshot/file as a source description. Never treat an unverified screenshot as an official source.

If online research is unavailable, use the user-provided JD as the primary source, mark source confidence as limited, ask for an official link if needed, and do not invent missing requirements.

Confirm the exact employer, title, location, seniority, requisition, URL, and posting status. Prefer current official employer sources; cross-check official team/product/ research pages and recent comparable roles. Record URLs, publication/access dates, source type, evidence level, and confidence.

Collect principal work activities, expected outputs, stakeholders, mandatory and preferred skills, methods, tools, standards, seniority, and logistical constraints. Separate explicit requirements, tentative requirements, and generic market signals.

3a. Complete publication and patent metadata when needed

When the user provides incomplete publication or patent information and the information is relevant to the resume, search online to complete verifiable metadata.

For publications, search by DOI, exact title, title + first author, journal + year + author, or publisher/Crossref/PubMed/Web of Science records when accessible.

For patents, search by application number, publication number, grant number, exact title, inventor + title keywords, CNIPA, Google Patents, WIPO PATENTSCOPE, USPTO, or other official patent databases.

Only complete fields supported by sources. If sources conflict, preserve the conflict and mark the field 需用户确认.

4. Evaluate fit and select evidence

Normalize sources and extract requirements using the module schemas. Convert the master profile into atomic candidate evidence. Follow [references/matchingscoringrules.md](references/matchingscoringrules.md) and [references/jobmatching.md](references/jobmatching.md). Map each requirement to candidate evidence IDs, calculate the 0-5 match score, requirement weight, credibility coefficient, and composite score, then identify eligibility or credibility risks. Flag but do not automatically include low GPA, failed courses, unfinished degrees, unexplained gaps, unrelated short experiences, weak low-relevance awards or certificates, unverified claims, exaggerated proficiency, sensitive personal information, or publications listed as accepted/submitted/in preparation without evidence. Omit when unnecessary, reframe truthfully when relevant, and mark 需用户确认 when uncertain.

Rank projects and roles, then classify them as:

• lead: early placement and detailed bullets;
• support: concise complementary evidence;
• omit: retained in the master profile but excluded from this resume.

Use the adoption thresholds from [references/experienceselectionrules.md](references/experienceselectionrules.md): lead requires a high-priority requirement with composite score >= 16; support covers composite score 8-15 or useful complementary evidence; omit covers composite score \ --template --out

For HTML/PDF validation:

```bash
python scripts/check_resume.py --html 
python scripts/render_pdf.py --in  --out  --paper A4

Inspect the rendered PDF visually.

If PDF rendering fails, provide HTML and Markdown, report the validation or rendering failure, and do not claim the PDF was generated.

8. Handle photo requests

Read references/photo_generation_rules.md only when the user asks for resume/profile photo advice together with a resume task. If the user supplies their own image and requests editing, use image generation/editing tooling with the closest prompt in assets/photo_prompts/.

If no personal photo is supplied, provide a prompt and request an upload; do not generate a specific fake identity. State that generated/edited images are for resumes, professional profiles, or internal display, not official identity documents.

9. Quality check and deliver

Run references/output_quality_checklist.md. Report:

• 信息完整度诊断;
• 缺失信息清单;
• 简历定位与岗位匹配建议;
• 简历正文或文件;
• 可选优化版本;
• 求职照片建议与提示词 when relevant and explicitly requested with a resume task;
• 下一步优化建议.

For targeted applications, present:

• job-analysis-report.md;
• matching-matrix.md;
• targeted-resume.md or targeted-resume.html;
• missing-info.md;

Only add content tuning sheet, evidence trace table, bilingual version, photo prompt, or full master profile JSON when explicitly requested.

Omit irrelevant sections instead of outputting empty headings.

Measurable Quality Gates

Length and Density

For one-page Chinese resumes:

• recommended total length: 700-1200 Chinese characters, excluding contact information;
• each proj

…

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: LittleDrunkWang
• Source: LittleDrunkWang/resume-builder-cn
• License: MIT

Install and usage instructions live in the source repository linked above.