site2voice 0.5.1

Ready-to-use VOICE.md files and optional benchmarks for AI-generated copy.

site2voice

Drop-in VOICE.md files. Pick a website voice, save it in your repo, and your agent writes less generic copy.

DESIGN.md tells agents how UI should look. VOICE.md tells them how copy should sound.

No install. No JSON. No generation step.

Voices

Voice	Tone	Best for
Apple	premium, minimal	product launches
Anthropic	measured, institutional	AI safety and product pages
EYESMAG	concise, culture-led	Korean fashion and lifestyle copy
Figma	collaborative, creative	design-tool pages
GitHub	developer-first	platform and ecosystem copy
Highsnobiety	editorial, commerce-aware	culture and fashion launches
Hypebeast	trend-led	streetwear and culture blurbs
Linear	precise, product-team	SaaS positioning
Monocle	polished, global	city, design, and affairs copy
Notion	simple, workspace	productivity pages
OpenAI	research-to-product	AI product pages
Shopify	merchant-growth	commerce pages
Stripe	calm, technical	fintech and SaaS copy
Vercel	performance-led	developer platform pages
Wallpaper	design-editorial	architecture and interiors copy

Use

Download one voice as VOICE.md:

curl -L https://raw.githubusercontent.com/SihyeonJeon/site2voice/main/voices/stripe.md -o VOICE.md

Tell Claude Code, Codex, Cursor, or Copilot:

Use @VOICE.md for headings, CTAs, navigation labels, and UI microcopy.

Each file is a plain Markdown writing brief with sentence rhythm, heading shape, CTA verbs, content boundaries, claim boundaries, and a benchmark target.

These are not official brand guidelines. They are derived style briefs from public pages, built to help agents write nearby copy without pasting source prose.

Before / After

Same LedgerFlow prompt, scored against Stripe:

Candidate	Result	Overall	Variety	CTA	Copy safety
Without VOICE.md	FAIL	60.4	40.5	50.0	100.0
With VOICE.md	PASS	91.0	45.8	100.0	96.2

See the full comparison.

Optional CLI

Generate a new voice from any public URL:

pipx install site2voice
site2voice init https://example.com

Validate generated copy:

site2voice bench https://example.com draft.md --strict

Need voice.json or an agent prompt too? Use the full context packs.

Status

Develop

python3 -m pip install -e .
make test
make bench-ci

License

MIT