Resume Generator

Generate and tailor one-page resume PDFs from RenderCV YAML.

The package provides two CLI paths:

• Render an existing RenderCV YAML file to PDF.
• Tailor a resume from a public job URL and local Markdown source material, then validate generated claims before rendering.

Install

uv tool install evidence-resume-tailor

For local development:

uv sync

Render A Resume

uv run resume-generator examples/basic.yaml

The PDF is written to:

output/resumes/basic.pdf

RenderCV also writes a .typ build artifact next to the PDF. Generated resumes are capped at one page; oversized PDFs fail the command so the resume stays targeted. When local PDF rasterization is available, the CLI also reports page fill, for example 88% page height reached.

The render command can also be called explicitly:

uv run resume-generator render examples/basic.yaml --output-dir output/resumes

Use --keep-supporting-formats when you also want RenderCV's Markdown and HTML outputs.

Tailor A Resume

The tailoring flow uses local Markdown files as the source of truth, fetches a public job page, asks Codex CLI to produce structured tailored resume content, validates generated resume claims against cited source text, then renders the final one-page PDF with RenderCV.

uv run resume-generator tailor \
  --job-url "https://company.com/jobs/123" \
  --content-dir content \
  --out-dir generated/company-role

Artifacts are written under the output directory:

job.md
prompt.md
tailored_resume.json
resume.yaml
evidence.json
resume.pdf
resume.typ

Each run writes to a staging directory first. Existing final artifacts are only replaced after evidence validation and one-page PDF rendering both succeed. If a valid one-page draft uses too little vertical space, the tailor flow gives Codex one bounded revision attempt with the measured page-use percentage and asks it to add only stronger job-matched, evidence-backed content. If a run fails after writing diagnostics, the CLI reports the staging path.

Tailoring requires the codex CLI to be installed and authenticated. Plain YAML rendering does not require Codex.

Local Source Material

Put private source material in Markdown files under content/. Nested folders are fine, and each file becomes an evidence source keyed by its relative path.

content/
  profile.md
  projects/
    resume-generator.md

Optional YAML frontmatter is supported:

---
kind: project
---

Built a RenderCV-based resume generator.

The AI is instructed to use only these files and to choose only the strongest material for the specific job. Substantive generated resume claims in summaries, highlights, bullet-style entries, section fields such as company/project names and dates, skill details, and cv.headline must be cited in evidence.json. Short display labels such as skill group names are treated as presentation, not claims. The validator also checks that cited source files contain the claim or the claim's key terms and numbers; otherwise the PDF render is blocked.

content/, generated/, output/, tmp/, and PDF/Typst render artifacts are ignored by git so personal material stays local by default.

Development

uv run pytest
uv build --wheel --sdist

To verify the local Codex CLI integration:

RUN_LIVE_CODEX_SMOKE=1 uv run pytest tests/test_codex_smoke.py

Release Automation

The repository includes a GitHub Actions workflow that can publish to PyPI through trusted publishing.

The scheduled publish workflow runs around 1:00 AM Eastern. It checks package-relevant files against the most recent v* tag. If no package changes are found, it exits without publishing. If package changes are found, it bumps the patch version, commits the version update, creates a tag, builds the wheel and source distribution, publishes to PyPI, and creates a GitHub release.

PyPI trusted publishing must be configured for this repository before the first automated publish can succeed. Configure the PyPI project with:

PyPI Project Name: evidence-resume-tailor
Owner: the-Drunken-coder
Repository name: resume-generator
Workflow filename: publish.yml
Environment name: pypi

If the PyPI project does not exist yet, create a pending trusted publisher with the same values in PyPI before running the workflow.