patina

Strip the AI packaging. Keep the meaning.

한국어 · 中文 · 日本語 · English

Try it in the browser — no install

Paste AI-sounding text → patina rewrites it in place, keeps the facts (the "30 templates" number survives), and drops the deterministic AI signal 100 → 0 (MPS 100 / Fidelity 75).

patina is a deterministic, pattern-based humanizer for Korean, English, Chinese, and Japanese. It finds AI-sounding phrasing and rewrites it without changing the claim, numbers, polarity, or causation — built for allowed AI-assisted drafting, not for evading detectors.

• Auditable, not a black box — 184 named patterns drive every edit; --diff shows exactly what changed and why.
• Meaning survives, verified — every rewrite is gated by meaning-preservation (MPS) and fidelity floors; drifted rewrites are retried or rolled back.
• Three independent axes — Document Type owns genre, Persona owns voice, Register owns delivery. Omit any axis to preserve the source.
• Every surface — agent skill (Claude Code · Codex · Cursor · OpenCode), Node CLI, and a browser playground.
• Honest about limits — scores are editing signals, not authorship proof; our own pre-registered study publishes where rewriting fails alongside where it works.

Quick Start

Browser — nothing to install. Open patina.vibetip.help and paste text. Rewrites run server-side with the MPS/fidelity gates; API mode forwards your own key per request (never stored or logged).

Agent skill — paste this into Claude Code, Codex CLI, Cursor, or any agent:

Install patina by following https://raw.githubusercontent.com/devswha/patina/main/INSTALLATION.md

Then use it:

/patina --lang en

[paste your text here]

CLI — Node >= 18:

npx patina-cli --lang en input.txt          # rewrite
npx patina-cli doctor                       # check backends and keys

A logged-in codex, claude, or gemini CLI works with no API key: add --backend codex-cli. Full install options: INSTALLATION.md.

Three Independent Axes

patina does not infer one axis from another. Omit Persona and Register to keep the source voice and register.

Axis	Controls	Does not control	Select with
Document Type	Genre, purpose, structural conventions, pattern policy	Voice, casual/professional delivery, meaning floors	--document-type · config document-type · Playground "Document Type"
Persona	Reusable voice fingerprint: vocabulary, rhythm, explanation habits	Genre, pattern policy, register, meaning floors	--persona · config persona · Playground "Persona"
Register	casual or professional delivery	Genre, persona identity, pattern policy	--register · config register · Playground "Register"

Meaning preservation is the outer guard; an explicit value never fills an omitted axis.

Commands

patina input.txt                                          # rewrite with defaults
patina --audit input.txt                                  # detect patterns only
patina --score --offline --exit-on 30 input.txt           # deterministic CI gate, no API key
patina --diff input.txt                                   # pattern-by-pattern changes
patina --verify input.txt                                 # rewrite + MPS/fidelity floor check
patina --document-type email --register professional input.txt
patina persona new my-voice --from-sample past-posts.txt  # learn a reusable voice
patina --persona my-voice draft.md
patina --batch docs/*.md --outdir cleaned/

patina --help prints the full flag list. CI wrapper for GitHub Actions: devswha/patina-action — plus pre-commit, static-site, Docker, and release integrations.

Project config lives in .patina.yaml:

version: "7.0.0"
language: ko              # ko | en | zh | ja
document-type: default    # genre/purpose + pattern policy
persona:                  # optional reusable voice; omit to preserve source
register:                 # casual | professional; omit to preserve source

Facts

184 patterns	37 rewrite-capable + 9 score-only viral-hook per language (46 each across KO/EN/ZH/JA) — see the full 184-pattern catalog in PATTERNS.md
Modes	rewrite · verify · audit · score · diff
Calibration	67.3% editing-hotspot catch [63.5–71.0%] across GPT-5.5 / Claude Sonnet 4.6 / Gemini 2.5 Pro (n=600, KO+EN); 16.0% false positives [11.6–21.7%] on KO+EN human controls (n=200)
License	MIT

Scores are editing signals with false positives and false negatives, not proof of authorship. See Ethics.

Documentation

• Cookbook — common recipes · CLI contract — flags, gates, exit codes
• Before/After gallery (한국어) · Pattern catalog
• Architecture · Configuration & authentication
• Benchmarks · Research · FAQ (한국어)
• Contributing (한국어) · Changelog

License

MIT. See LICENSE and NOTICE. Inspired by oh-my-zsh, Wikipedia's "Signs of AI writing", and blader/humanizer.