Cv builder

CV Builder

Overview

Turn scattered developer experience — local projects, GitHub repositories, old resume files, or plain-text notes — into a polished, ready-to-submit resume or CV. This skill acts as a technical recruiter: read source materials in parallel, ask targeted follow-up questions, draft content in Markdown, and render it to HTML/PDF via customizable templates.

When to Use

Use when the user wants to:

• Create a new resume or CV from scratch
• Turn GitHub projects or local codebases into resume bullet points
• Update or reformat an existing resume/CV
• Generate a PDF resume with a specific visual style
• Tailor a resume for a specific role or industry

When Not to Use

Do not use for:

• Writing cover letters or application emails
• Submitting resumes to job boards or company portals
• Verifying the truthfulness of work history or project claims
• Guaranteeing ATS compatibility or interview responses
• Non-technical resumes unless the user explicitly provides all content

Boundaries and Defaults

Can do: read local projects, GitHub repos, old resume files, or plain-text notes; dispatch sub-agents to read materials in parallel; ask targeted follow-up questions; draft a Markdown resume; render to HTML/PDF via built-in (modern/classic/minimal) or custom templates.

Cannot do (without explicit user approval): fabricate experience, projects, or metrics not in the source materials; submit resumes to job boards or company portals; modify any source files outside the skill's output directory.

Default behavior: read-only on source materials; all writes go to the user's chosen output directory and require approval; the user confirms every personal claim (employment dates, project metrics, technologies used) before it enters the draft.

Core Workflow

Execute the following steps in order. During interactive phases where the user must make choices, use the structured question/interaction capabilities provided by the current Agent (e.g., multiple choice, single choice, confirmation boxes) instead of dumping large blocks of text questions into the chat.

Step 1: Collect Input Sources

Ask the user what materials they have. Any combination is supported:

Source	Handling
Local project folder	Read README, package.json, pyproject.toml, and source-code summaries
GitHub repository URL	Fetch repository metadata and README
Existing resume file (PDF/DOCX/MD)	Extract text; on macOS, use textutil -convert txt for DOCX
Plain text or notes	Accept as-is
LinkedIn / Notion / portfolio URL	Scrape if possible; otherwise ask the user to export

Do not start generating content until the user has provided all materials.

Step 2: Dispatch Sub Agents to Read Materials

The main agent decides the dispatch strategy based on project size:

• Large projects (many files, complex code): assign a dedicated sub agent
• Small projects or simple files: combine into one sub agent
• Old resume / text notes: one sub agent

Each sub agent returns a concise summary including:

• Project/file purpose
• Key technologies used
• Notable features or achievements
• Quantifiable impact (if any)
• Suggested resume bullet points

Step 3: Ask for Personal Information

Confirm or collect the following through structured interactions, grouped by topic:

1. Basic info: name, email, phone, location, LinkedIn, GitHub
2. Career goal: target role, industry, city
3. Experience and skills: education, work experience, projects, tech stack
4. Bonus items: certifications, languages, awards, open-source contributions
5. Photo (optional): whether to provide an ID photo and place it in the top-right corner of the resume

Pre-fill answers using the source summaries and let the user edit or add details.

Step 4: Draft the Markdown Resume

Generate a Markdown resume that:

• Has a clear, ATS-friendly structure
• Includes keywords for the target role
• Uses concise bullet points (preferably STAR method)
• Matches skills to the target role

Show the draft to the user and ask them to edit it. Repeat until the user approves the content.

Step 5: Choose a Template

Ask the user for their preferred style through structured interactions:

• Built-in templates: modern (modern single-column), classic (classic two-column), minimal (minimal)
• User-provided template: accept an HTML/CSS file path
• Agent-designed: user describes the style, and the agent generates HTML/CSS

Render an HTML preview and show it to the user.

Step 6: Confirm Page Layout

Ask the user about layout preferences through structured interactions:

• Page count: squeeze to one page or allow multiple pages?
• Photo: provide an ID photo and place it in the top-right corner?
• Other adjustments: font size, line spacing, margins, etc.?

Modify the HTML/CSS and re-render the preview based on the user's choices. If the user chooses one page, use build_resume.py --one-page to generate a compact layout.

Step 7: Generate the PDF

Convert the HTML preview to PDF using WeasyPrint.

Dependency handling:

1. Check whether a dedicated uv virtual environment exists in the skill directory and whether WeasyPrint is installed
2. If not, create the environment and run uv pip install weasyprint
3. If automatic installation fails, output the HTML preview and provide manual installation instructions

Iterate with the user on layout, content, and template until satisfied.

Resources

• scripts/build_resume.py — main entry point (supports --one-page compact layout)
• scripts/render_pdf.py — HTML to PDF (WeasyPrint, auto-injects Hiragino Sans GB font on macOS)
• scripts/ensure_weasyprint.py — manages the uv environment
• scripts/read_project.py — scans key files in local projects
• assets/templates/ — built-in HTML/CSS templates
• references/example-resume.md — example Markdown resume format