cv-claw

Hand Claude a PDF, a screenshot, or a job posting and ask in plain language — "turn this into a resume," "tailor it for this role," "make the name bigger" — and get back a polished, print-ready HTML document you can open and export to PDF. Keep your resume working with Claude instead of fighting a document editor.

The problem. The common advice is to tailor your resume to each job description — but maintaining that by hand is a pain. The template generally stays the same; what changes is the content that goes into it. AI chat can rewrite the words and style them — but each conversation is a one-off: a fresh document, regenerated from scratch, with no canonical original behind it. Run it again for the next job and the styling drifts; tailor five roles and you're juggling five unrelated files with no shared source of truth. There's no clean separation between what your resume says and how it looks, so the two can't vary independently.

What cv-claw does. It splits a resume into structured content (JSON, validated against a fixed schema) and a visual template (Jinja2 + CSS), then renders the two into a single self-contained HTML file — no browser automation, no dev server, no frontend toolchain. Structured content is what makes the resume editable by an agent: Claude works on JSON and templates, then runs one command to produce the artifact. You talk to Claude; cv-claw renders.

The template side is just as open. Because it's plain Jinja2 + CSS that Claude writes, there's no design barrier — you're not picking from a fixed gallery. Describe the look you want, or hand over a reference, and Claude authors the template to match — and because content and template are separate, swapping the layout never touches a word of your resume. Any layout you can describe is a layout you can render.

📦 Install

cv-claw is a standalone CLI. Install it however you manage Python tools:

# pip
pip install cv-claw

# uv (installs the CLI on your PATH, isolated)
uv tool install cv-claw

Either way you get the cv-claw command. To enable the optional PDF backend, install the corresponding extra:

pip install 'cv-claw[pdf]'     # native PDF export

🚀 Quickstart

cv-claw is built to be driven by Claude. The fastest path is to install the CLI and the skill, then describe what you want in plain language — Claude produces the JSON and runs the render for you. Start here.

Use it through Claude

The skills/cv-claw/ directory ships a single Agent Skill covering four resume tasks — ingest (PDF/image/text → JSON), tailor (adapt for a job description), tweak-template (restyle the current render), and create-template (new Jinja2 layout). The main SKILL.md stays small; each task lives under references/ and is loaded on demand.

Claude Code (recommended)

Every release attaches a packaged cv-claw-skill.zip. The link below always points at the newest release:

# Install the CLI
pip install cv-claw   # or: uv tool install cv-claw

# Drop the skill into Claude's skills directory
mkdir -p ~/.claude/skills
curl -fsSL -o /tmp/cv-claw-skill.zip \
  https://github.com/farhan0167/cv-claw/releases/latest/download/cv-claw-skill.zip
unzip -o /tmp/cv-claw-skill.zip -d ~/.claude/skills

This leaves you with ~/.claude/skills/cv-claw/. Point your agent at it and start producing resume JSON. Substitute a project-local .claude/skills/ for the home directory if you want it scoped to one workspace.

claude.ai and Claude Desktop

Download the same cv-claw-skill.zip and import it through the Skills UI — no CLI or unzipping needed. Open Settings → Capabilities → Skills, choose Upload skill, and select the zip.

From a clone

If you've cloned this repo, copy skills/cv-claw/ into your workspace's skills directory directly — it's the same content the release zip is built from.

Drive the CLI by hand

If you'd rather work without the skill, write the bundled example resume to start from, then render it:

cv-claw init                  # → writes resume.json (a complete example)
cv-claw render resume.json    # → writes resume.html next to the JSON

cv-claw init won't clobber an existing file — pass a path or --force (cv-claw init my-resume.json). The example is a full, realistic resume exercising every section kind; edit it into your own and re-render.

cv-claw renders whichever path you give it — the JSON can live anywhere. To keep generated HTML out of the way:

cv-claw render path/to/resume.json --output-dir build/
# → writes build/resume.html

Open the HTML in a browser and use the browser's print dialog to export to PDF. classic is the only bundled template; run cv-claw list-templates to see what's available.

⚙️ CLI

init

cv-claw init [path] [flags]

Write the bundled example resume JSON to path (default resume.json). Refuses to overwrite an existing file unless --force is given. Parent directories are created as needed.

Flag	Default	Description
[path]	resume.json	Destination file for the example JSON.
--force	off	Overwrite an existing file at the destination.

render

cv-claw render <input.json> [flags]

Validate the JSON, look up the template, and write a standalone HTML file. By default the output lands next to the input with an .html suffix.

Flag	Default	Description
-o, --output <path>	<input>.html	Single explicit output file. Wins over --output-dir.
--output-dir <dir>	—	Output directory; filename is <input-stem>.html. Directory is created if missing.
--template <name>	from JSON	Override the template field in the resume JSON for this render only.
--templates-dir <dir>	auto-discover	Replace the default search roots (workspace + bundled) with a single directory.

list-templates

cv-claw list-templates [flags]

Print discovered template names, one per line. Default output annotates each entry with its source: (workspace) for templates under ./.cvclaw/templates/, (bundled) for those shipped inside cv-claw.

Flag	Default	Description
--templates-dir <dir>	auto-discover	Replace the default search roots with a single directory; drops the source annotation.

validate

cv-claw validate <input.json>

Run schema validation only. Exits 0 on success, non-zero on failure. Useful as a sanity check before rendering.

Global

Flag	Description
--version	Show version and exit.

🧬 Schema

A resume is a JSON document with a template, a header, and a list of sections. Each section is one of four kinds — prose, keyvalue, list, or timeline — and is dispatched to the matching renderer in the template.

🎨 Templates

Templates live in two roots, both auto-discovered:

1. Bundled — ship inside the cv-claw package (e.g. classic). Read-only; they come down with pip install cv-claw.
2. Workspace — under ./.cvclaw/templates/<name>/ in your current working directory. Optional; created on first use by the skill.

A workspace template shadows a bundled template of the same name — that's the supported way to fork. The .cvclaw/ prefix is reserved for cv-claw workspace state; today it holds templates/, and future versions may add cache or config under the same prefix.

A typical workspace tree:

my-resume-workspace/
├── resumes/                       # or wherever you keep your JSON
│   └── default.json
└── .cvclaw/
    └── templates/
        └── minimalist/
            ├── minimalist.html.j2 # entry point
            ├── minimalist.css     # inlined into the rendered HTML
            └── _macros.html.j2    # optional section renderers

Files and directories starting with _ are treated as partials and are skipped by template discovery. Add a new template by dropping in a new folder with the same shape (the create-template task automates the scaffolding).

📄 Resume JSON conventions

cv-claw doesn't impose a directory for your resume JSON — pass any path on the CLI and it just works. The bundled skill (used by Claude Code and other agent surfaces) asks the user once where resume JSON should live and can record the answer in your workspace CLAUDE.md under a ## cv-claw: resume location heading so future sessions don't re-ask.

🛠️ Development

make install     # uv sync --all-extras
make lint        # ruff check
make format      # ruff format
make check       # ruff check + ruff format --check
make render      # render resumes/example.json
make clean       # remove caches and build artifacts

📜 License

MIT — see LICENSE.