resume-tailor

name	resume-tailor
description	Use when user provides JD and existing resume, expecting job-targeted optimization, ATS keyword alignment, language polishing, and delivery of a single-page A4 PDF resume.

Resume Tailor Skill

Core Objectives

• ATS Hit Rate: Cover high-frequency JD keywords while maintaining natural expression.
• Job Relevance: Prioritize evidence chains and quantified results that best match the position.
• Delivery Quality: Output A4 PDF with extractable text and pass format quality checks.

Inviolable Principles

• No Fabrication: Only rewrite, rearrange, compress—no fictional experience/tech/data.
• Volume Gate Before Generation: Compress content to single-page target before generating PDF.
• Autonomous Decision Making: Agent autonomously makes all optimization decisions (keyword alignment, content prioritization, compression, layout tuning) without asking for confirmation.
• Transparent Reporting: All decisions are recorded in the final summary report for user post-hoc review.
• ATS Friendly: No table-based main layout, no images replacing body text, no key info in headers/footers.
• Fixed A4 + Single Page: Default delivery 210mm x 297mm, 1 page.
• PDF Only Delivery: Final output is PDF only.
• Read vendor/skills/pdf/SKILL.md before PDF actions: Execute before initial generation, minor regeneration, or final export.
• Auto-fit Scope: Automatic tuning may adjust layout parameters only; it must not rewrite resume content unless user explicitly asks.

Stateless Boundary (Mandatory)

• resume-tailor skill directory can only store "rules + templates + scripts + references".
• No personalized cache (user preferences, historical JDs, contact info, resume samples) in skill directory.
• Unified directories (relative to workspace):
  • resume_output/: Latest PDF
  • resume_output/backup/: Historical PDF backups
  • cache/user-profile.md: Long-term preference cache
  • cache/resume-working.json: Current session resume body
  • cache/jd-analysis.json: JD analysis results (keywords, alignment, optimization actions)

Minimum Execution Flow

Phase A – Initialize

1. Run scripts/resume_cache_manager.py reset.
2. Run template-check; if template exists, run template-use to load into working cache. If not, run template-init with user-provided resume, then template-use.

Phase B – Analyze & Draft

1. JD Diagnosis: Analyze JD (or target direction), produce P1/P2/P3 tier classification with gap report, and persist results to cache/jd-analysis.json via jd-save.
2. Apply All Modifications: Apply optimization decisions using standard action codes (LEAD_WITH, EMPHASIZE, QUANTIFY, DOWNPLAY, MERGE, REWORD) as defined in references/optimization-actions.md. Document each action with target, code, and reason. Then run update to persist.

Phase C – Compress & Quality

1. Volume Gate: Score bullets via score_all_bullets() against cache/jd-analysis.json, then check working cache against volume thresholds; if compression needed, prioritize removing lowest-scored bullets first, then update.
2. QA & De-AI: Read vendor/skills/humanizer/SKILL.md for natural expression guidance, then run structure/quantification/ATS checks.

Phase D – Generate & Deliver

1. PDF Generation: Read vendor/skills/pdf/SKILL.md, then generate with --auto-fit. If QC fails, retry up to 3 times with escalating parameters.
2. Quality Report: Run python3 -m scripts.generate_quality_report --resume cache/resume-working.json --jd-analysis cache/jd-analysis.json --pdf resume_output/resume.pdf and embed the stdout output verbatim in the reply.
3. Summary Report: Output a structured summary report covering all decisions made. Must include the full absolute path of the generated PDF file.
4. Wrap-up: Update cache/user-profile.md, retain working cache.

Complete checklist and thresholds in references/execution-checklist.md.

Summary Report Format

After PDF delivery, output the following report:

## Resume Tailor Summary Report

### Job Analysis (from cache/jd-analysis.json)
- Position: {title}
- P1 (Critical): {keywords}
- P2 (Important): {keywords}
- P3 (Nice-to-have): {keywords}
- Matched: {aligned keywords}
- Gaps: {gap keywords with strategies}

### Modifications Applied
- Actions applied: {list of action_code → target, e.g., EMPHASIZE → experience[0].bullets[2]}
- Keywords added: {list}
- Content reordered: {yes/no, brief description}
- Content removed: {list of removed items with reasons}

### Volume & Compression
- Before: {word count} words / {line count} lines / {bullet count} bullets
- After: {word count} words / {line count} lines / {bullet count} bullets
- Compression actions: {list}

### QA Results
- Humanizer: {applied / not needed}
- Structure check: {pass/fail + details}
- ATS keyword coverage: {percentage or list}
- PDF QC: {pass/fail + checks summary}

### Deliverable
- PDF path: {full absolute path of generated PDF}

### Action Required
- Review contact info and sensitive data in the generated PDF.
- {Any failed QC checks or caveats, if applicable}

Fixed Template Baseline (Logical Structure)

Always organize content in the following module order:

Header
Summary
Professional Experience
Technical Skills
Education

Script Responsibility Boundaries

• scripts/resume_cache_manager.py: Manage cache/resume-working.json reset/init/update/show (cleanup only manual as needed), cache/base-resume.json template-init/template-use/template-show/template-check, and cache/jd-analysis.json jd-save/jd-show.
• scripts/generate_final_resume.py: Accept --input-json, optionally --auto-fit, and generate final PDF.
• scripts/generate_quality_report.py: Print quality report (Markdown) to stdout — no files written. Covers format compliance (PDF), content quality checks, keyword→bullet coverage matrix, and strategy summary.
• scripts/check_pdf_quality.py: Perform general format and text quality checks.
• scripts/check_content_quality.py: Content-level quality checks on resume JSON — bullet scoring, verb strength, quantification rate.
• scripts/resume_shared.py: Shared utilities (validation, JSON I/O, parsing helpers) consumed by other scripts.
• templates/modern_resume_template.py: Only responsible for PDF layout and export.

Bundled Skills (vendor/)

All dependent skills are bundled in the vendor/skills/ directory — no separate installation required.

• vendor/skills/pdf/SKILL.md: PDF read/generate guide — read before any PDF action
• vendor/skills/docx/SKILL.md: DOCX read/edit guide — read before reading .docx files
• vendor/skills/humanizer/SKILL.md: AI trace removal guide — read before humanizing text

Reference Materials

• Full process checklist and thresholds: references/execution-checklist.md
• ATS strategy: references/ats-keywords-strategy.md
• Prompt templates: references/prompt-recipes.md
• Profile cache template: references/profile-cache-template.md
• Working cache structure: references/resume-working-schema.md
• Template instructions: templates/README.md
• Optimization action codes: references/optimization-actions.md