Six Seconds Resume

Live at https://sixseconds.tech — a free resume builder that runs in your browser: build a resume, check one, templates, examples and guides.

A recruiter spends about six seconds on a resume before deciding. Everything here — one column, no decoration, parseable by machine first — exists to survive that.

Build a single-column resume and download it as PDF, DOCX, and plain text. The builder runs entirely in the browser — no account needed, no server-side rendering of your document, nothing uploaded. An optional passwordless account syncs resumes between devices; nothing is sent to the server until you save it there.

Start from the resume you already have. Drop a PDF or a Word .docx into the builder and it is parsed into an editable document — in your browser, with nothing uploaded — and the import lands in a review that says, field by field, how sure the parser is of what it read. Ctrl+Z puts your previous draft back.

/check — see what a machine reads. The same extraction, pointed at any resume, with no account and no signup wall: the text a parser recovers, the fields it finds, and the lines where two different parsers would disagree about reading order. It describes what was recovered and never claims what an employer's system will do with it, because nobody can know that.

Twenty-four templates, and none of them changes what a parser reads. They differ in typeface, header placement, heading style, paper size, density and section order — including presets shaped by where you are applying (Letter and one page for the US and Canada, A4 and two for the UK, room for three in Australia) and by what you do (a licence read before a job history in nursing, a publication list that is the work rather than a footnote in research). Every one stays a single column of real text with standard headings, and a test asserts that switching between them leaves the extracted text byte-identical — /templates shows them as real rendered pages.

A phrase library and a bullet coach, with no model anywhere. The library inserts shapes with the blanks still in them (Cut ___ from ___ to ___), so nothing it gives you is a claim you did not make. The coach decomposes a bullet into action, what, how and outcome, and asks about the part that is missing — it never writes the sentence. Both are constrained by D8, and the whole phrase bank is asserted to pass the app's own lint engine.

Examples and guides. /examples publishes eight complete resumes with the plain text a parser recovers from each as the page body, and /guides has four written to answer their question rather than to exist.

Getting started

pnpm install
cp .env.example .env   # then set AUTH_SECRET
pnpm db:migrate
pnpm dev

Then open http://localhost:3000.

.env.example documents every variable. Only two matter for local development: DATABASE_URL and AUTH_SECRET. With AUTH_DEV_OUTBOX set (it is, in the example), sign-in emails are written to that directory as JSON instead of being sent, so the whole magic-link flow works without an email account. That transport refuses to run under NODE_ENV=production; a real deploy needs EMAIL_SERVER.

predev and prebuild run scripts/sync-public-fonts.mjs, which copies the vendored font files and pdfjs's worker into public/. Both are generated and gitignored — the ~2.8MB of fonts lives in the repository exactly once, under src/lib/fonts/files/. If the preview 404s on a font, run pnpm fonts:sync.

Scripts

For a read-only check of every public sitemap URL, run node scripts/audit-seo.mjs https://sixseconds.tech report.json. It checks rendered metadata, indexability, structured-data JSON and internal discovery links, and exits nonzero on failures. Actual Google positions and indexing status come from Search Console; see docs/SEO.md.

Command	What it does
pnpm dev	Development server
pnpm build	Production build (standalone output, for Docker)
pnpm verify	typecheck + lint + test — run before committing
pnpm test	Vitest unit and component suite
pnpm test:e2e	Playwright end-to-end suite (builds and starts the app)
pnpm fonts:fetch	Re-vendors the font files from Google Fonts (rarely needed)
pnpm fonts:sync	Copies fonts and the pdfjs worker into public/
pnpm auth:google	Asks Google whether the OAuth credentials and redirect URI it has match this deployment
pnpm skills:build	Rebuilds data/skills.json from O*NET (needs network; the output is committed)
pnpm phrases:build	Rebuilds data/phrases.json from O*NET (needs network; the output is committed)

Run pnpm test:e2e before trusting anything about the preview. The PDF is generated in a Web Worker and painted with pdfjs, and pdfjs behaves differently under Node than in a browser — a whole class of bug there is invisible to the unit suite.

Deployment

Vercel cannot host this. The architecture assumes a persistent filesystem, which Vercel's ephemeral functions do not provide. Next.js defaults push everyone toward Vercel, so this is worth stating plainly.

This is the beginning of the README. Read the rest at the original. Read the full README ↗