I build small tools that do one thing, deterministically, on your own machine — no API keys, no accounts, nothing uploaded. Most of them started as "I need this and the only option is a website I'd have to paste a live token into."

Four lines of work:

Day job side: React / Next.js / TypeScript, and making sure what I ship is legible to search engines and answer engines alike (SEO · AEO · GEO).

const me = {
  building: ["agent tooling", "local-first tools", "small web products"],
  values:   ["deterministic", "zero-dependency", "nothing uploaded"],
  stack:    ["TypeScript", "React", "Next.js", "Node"],
} as const

_{The newest thing I work on, and the part I'd point at first.}

• oh-my-fable — a harness for long agent tasks: plan first, self-correct, and never lose the thread when the process dies halfway through.
• fable-audit — an autonomous codebase audit that survives crashes. Plans the audit, reads your actual files, resumes where it stopped.
• oh-my-mcp — lint your MCP server before a model chokes on it. Catches tool definitions that read fine to a human and terribly to an LLM.
• korean-patent-mcp — MCP server that answers whether a Korean patent is still in force, from KIPRIS Plus. Three tools, no guessing.
• ai-readme-generator — reads a codebase and writes the README it should have had.

The most convenient dev tools are websites you paste sensitive data into. These are the same tools, rebuilt to run entirely in your tab.

Security & privacy · csp-doctor (CSP XSS holes) · cookie-doctor (Set-Cookie flags) · scrubpix (strip EXIF/GPS) · unspook (invisible / Trojan-Source / homoglyph chars) · entropass (passwords + strength) · checkdigit (Luhn / IBAN / ISBN)

Text, color & media · is-this-ai-slop (AI-writing tells, with zero AI) · readlevel (readability) · spellnum (numbers → words, EN/KO) · huebrew (image → palette) · dichroma (color-blindness simulation) · captionkit (subtitles) · calforge (.ics + add-to-calendar)

Things I made because I wanted the answer · fake-star-detector (are those stars bought?) · wiki-beef (the pettiest Wikipedia edit wars, live) · is-this-job-toxic · is-this-passive-aggressive · red-flag-green-flag

_{TypeScript · zero runtime deps in the core · CLI + library · ESM + CJS · MIT.}

Linters that earn their place in CI · a11ylint (WCAG) · aeolint (SEO/AEO + rich-result eligibility) · apilint (OpenAPI) · datalint (CSV/TSV quality) · envlint (.env secrets & schema) · i18nlint (translation files) · licenselint (license audit + SBOM) · linklint (dead links & anchors) · maillint (HTML email compatibility) · docsanity (docs-site health) · actionbudget (CI-minute waste) · tsconfig-doctor (strictness & footguns) · will-it-redos (catastrophic regex)

Tiny libraries · pennywise (exact BigInt money) · nanocsv (CSV parser) · verspec (SemVer) · croniq (cron) · tempora (human time, ~1KB) · okcolor (OKLab/OKLCH + WCAG) · fuzzrank (fzf-quality matching) · fetchwise (retries/backoff/timeouts)

For fun, still deterministic · coderecap (Wrapped, for your git history) · shellrecap (Wrapped, for your shell) · works-on-my-machine (env fingerprint, paste-safe) · its-not-dns (it was DNS)

Apps in Toss mini apps and small web services, plus local-first tools built for a real habit rather than a demo:

• hunch — 판단을 기록하고 확률로 채점하는 결정 저널
• memoro — 개발자를 위한 간격 반복 학습
• netto — 청구한 돈이 실제로 들어왔는지 세금까지 맞춰 추적
• recur — 계좌를 연동하지 않는 정기 결제 관리

Find these useful? Sponsor on GitHub — it keeps them maintained.