/cv-optimize — Resume Analysis & Optimization

Read a resume file (PDF / Markdown / DOCX), perform structured analysis, and produce a target-role-optimized PDF.

Phase 0: Collect Input

1. Resume file: Read from user-provided path (supports PDF / Markdown)
2. Target info (from arguments or ask user):
   • Target role (e.g., "AI Engineer", "PM", "Data Scientist")
   • Target company name (optional, for tailoring)
   • JD content or link (optional, for keyword alignment)
   • Emphasis points (e.g., "leadership", "bilingual", "career transition")

If no target role is provided, ask once:

What role are you optimizing for? (e.g., AI Engineer / PM / Data Scientist / general)
Any specific strengths you want to emphasize?

Phase 1: Resume Analysis

After reading the full resume, output a structured analysis:

## Resume Analysis — {Name}

### Basic Info
- Name / Target title / Total experience / Education

### Strengths (Keep)
1. ... (list 3-5 highlights worth keeping, especially quantified achievements)

### Issues (Needs Optimization)
| # | Issue | Severity | Description |
|---|-------|----------|-------------|
| 1 | ... | High/Med/Low | ... |

### Target Role Match Analysis
| Requirement | Candidate's Experience | Match Level |
|-------------|----------------------|-------------|
| ... | ... | Strong/Medium/Weak/Gap |

Mandatory checks (every resume):

• Title vs target alignment
• Summary is focused (no more than 3 key positioning statements)
• Each experience has quantified results
• Skills section aligned with target role (flag outdated skills)
• Timeline is clear and consistent
• Structure depth doesn't exceed 2 levels
• Space allocation proportional to relevance
• Each role includes Tech Stack

Phase 2: Reposition & Optimize

2.1 Positioning Adjustment

• Redefine subtitle (role positioning statement)
• Rewrite Summary (focus on 2-3 key differentiators with numbers)
• If emphasis points were specified, highlight them in Summary and highlight box

2.2 Experience Restructuring

• Add Tech Stack tags to each role (blockquote or meta style)
• Expand highly relevant experiences with section-label sub-blocks
• Condense less relevant experiences to 2-4 bullets
• Use action + contribution + verified outcome. Quantify only from supplied evidence; otherwise use a factual qualitative result and mark missing metrics.
• Preserve the candidate's actual responsibility. Never turn participation or assistance into leadership without evidence.

2.3 Skills Section Rebuild

• Remove skills irrelevant to target role
• Group by category (e.g., AI/ML, Data, Infrastructure, Tools)
• Include target-role keywords only when supported by the candidate's actual experience. Mark gaps rather than inventing skills.

2.4 Special Sections (add when appropriate)

• "Why I'm right for X" highlight box: when target is clear and emphasis points exist
• "Career transition trajectory": when candidate has obvious career pivots
• Selected side projects table: keep only 3-5 relevant to target

Phase 3: Generate Optimized PDF

Check the existing PDF toolchain first. WeasyPrint requires Python and native rendering dependencies; it is not a zero-install guarantee. Ask before downloading packages or installing system dependencies.

3.1 HTML Structure

Element mapping:
- h1          → Name
- .subtitle   → Role positioning statement
- .contact    → Contact info
- h2          → Major section titles (Summary / Experience / Skills)
- .job-title  → Job title per role
- .job-company → Company + location + dates
- .job-meta   → Tech Stack (blue left-border blockquote)
- .section-label → Sub-section titles (gray background)
- ul > li     → Bullet points (single level only, no nesting)
- table       → Competencies / Skills / Projects
- .highlight-box → Special emphasis block (blue left-border + light gray background)

3.2 Layout Rules

• Target 2 pages (max 3, never leave a near-empty trailing page)
• Font size: headings 21-22pt / body 9-9.5pt / meta 8.5pt
• Line height 1.5-1.55
• Margins 16-18mm
• Table headers: dark blue background, white text; even rows light gray
• No 3+ level nesting — flatten deep structures

3.3 PDF Generation

# Run only in an approved environment with required native libraries
uv run --with weasyprint python3 script.py

After generation:

1. Use Read tool to verify PDF layout
2. If trailing page is too empty, adjust margin/font-size and regenerate
3. Confirm page count is reasonable, tell user the file path

3.4 Output File Naming

{original_file_directory}/{Name}_{Target_Role}_optimized.pdf

Phase 4: Change Summary

Output a concise comparison table:

| Item | Original | Optimized |
|------|----------|-----------|
| Positioning | ... | ... |
| Summary | ... | ... |
| Skills | ... | ... |
| Length | ... | ... |

Plus notes (e.g., sections where the candidate needs to provide specific numbers).

Design Principles

1. One resume, one narrative — all bullets serve the same positioning story
2. Quantify > describe — "improved 67%" beats "significantly improved" every time
3. Delete > shrink > keep — irrelevant experience: cut to 2 lines, don't waste half a page
4. ATS-friendly — ensure target role's core keywords appear in Summary + Skills + Experience
5. Clear visual hierarchy — 10-second scan reveals: name → positioning → 3 key numbers → timeline
6. No empty words — delete "eager learner", "team player", "value creator" — they carry zero information

Color System

Primary (dark blue):
- Heading text:      #0d1b2a
- Section titles:    #1b263b
- Accent:            #415a77
- Light background:  #f0f4f8
- Table header:      #1b263b (white text)
- Table even rows:   #f8f9fa
- Section labels:    #e8edf2

Quality Gates

• Dates, employers, degrees, responsibility and metrics trace to input evidence.
• Missing metrics remain questions or placeholders, never invented percentages.
• HTML escapes user content and uses local assets; no remote images, fonts or CSS.
• PDF generation does not fetch remote URLs or arbitrary local files referenced by input.
• Inspect layout AND extract text to check reading order; do not guarantee ATS compatibility.
• Preserve the original and use a new output path. DOCX requires a DOCX reader; do not claim direct Read support.