CV Tailor
CV Tailor
Personal · OSS · 2026

Overview

Ask an LLM to "tailor my CV for this job posting" and you get two failure modes at once: generic templates that ignore the job description, and confident fabrication — invented metrics, claimed-but-unused technologies, inflated titles. CV Tailor is an agentic skill for Claude Code (a SKILL.md pipeline any AI agent can execute) designed against both: it produces a genuinely tailored, ATS-friendly LaTeX CV where every bullet traces back to a source file you control.

The Pipeline

job description (URL or text)
        │
        ▼
1–3  fetch JD · read your resource library · extract seniority/stack/culture signals
        ▼
4    FIT ASSESSMENT — 3–5 strongest matches, 1–3 honest gaps,
        verdict (Qualified / Stretch / Overqualified) — shown BEFORE any drafting
        ▼
5–8  select bullets (ownership-first, compressed for older roles) · rewrite
        impact-first (XYZ: Outcome → Metric → Mechanism, ≤2 lines) · tailor
        summary to your voice · mirror JD keywords verbatim for ATS
        ▼
9–10  write Markdown + LaTeX · compile (tectonic → pdflatex → xelatex → latexmk)
        · debrief + archive to iCloud (per-company folders, aux files never sync)

The structural piece is the resource library: experiences live as brag files with structured sections — What I Did / Impact / Context / Framing Notes — where framing notes say which angle to emphasize per role type (the same CI/CD project reads as reliability for SRE, cost savings for FinOps). Selection is a real retrieval problem, not generation: the skill picks and re-frames what exists.

Honesty Rules (the actual feature)

  • No fabricated metrics — every number traces to resources/experiences/; unquantified impact stays descriptive
  • No claimed technologies you haven't used — missing skills surface as gaps in the fit assessment, not as skills
  • No title inflation — official titles stay; scope shows through ownership verbs
  • Pushes back — requests to add unverified tools or numbers are refused by design

Privacy & Tooling

  • Privacy by gitignore: the repo ships only *.template.md skeletons; real contact details, experience files, and brag notes never leave your machine via git
  • Templates: single-column ATS-friendly LaTeX with glyphtounicode (machine-readable PDF), 1–2 page discipline — cut bullets, never shrink fonts; dual-engine (pdfTeX/XeTeX compile identically)
  • Tectonic bootstrap: setup.sh installs a single static LaTeX binary (~30 MB, no sudo) — no TeX Live marathon

Tech stack

Tools Productivity

Architecture


            cv-tailor/
├── SKILL.md              # the skill: 10-step pipeline, modes, honesty rules
├── resources/
│   ├── meta/              # contact · about (voice) · location (templates only in git)
│   ├── educations/        # degrees, honors
│   └── experiences/       # brag files: What I Did / Impact / Context /
│                         #   Framing Notes + YAML meta (featured, overlap_risk)
├── templates/
│   └── ats-friendly-technical-resume/   # LaTeX + Markdown, dual-engine
├── scripts/               # setup.sh (tectonic) · compile.sh · publish-icloud.sh
└── output/                # generated CVs (gitignored → archived to iCloud)
…