I build local-first products, native Apple apps, and agent engineering systems.
My work focuses on useful software that respects personal data, explains its behavior, and can be built and verified repeatedly.
I develop a private product workspace with six Swift products:
| Product | What it does |
|---|---|
| Linkory | A personal link memory library and routing system: save, search, clean, repair, and open links in the right app. |
| Linko | A local-first iOS app for collecting and organizing social posts and web links with previews, tags, notes, and link health checks. |
| Hoppa | An iOS deep-link router with Action Extension handoff, Universal Link fallback, history, and diagnostics. |
| Luho／收行李啦 | A travel and packing planner for trips, people, checklists, schedules, documents, and attachments. |
| Minoto | A meeting recorder and notes app with live transcription, speaker/channel recognition, history, and optional BYOK summaries. |
| Swoop | A macOS window-management utility controlled by trackpad gestures, with tiling, grids, Spaces, and multiple-display support. |
stoat-keyboard is a native iOS Zhuyin, Dynamic Zhuyin, and Pinyin keyboard. It combines Swift, RIME/librime, Taiwan Traditional Chinese dictionaries, offline processing, and reproducible release builds.
- LINE Cheater — a Rust/Electron/Python toolkit for browsing, searching, exporting, validating, and slimming LINE backups while keeping source backups read-only.
- Baby Diary — a React + Vite mobile-first PWA for recording feeding, sleep, diapers, pumping, food, and growth with local or user-controlled GitHub storage.
- OysterunCore — a local host that provides a web chat interface for coding agents.
- claude-code-workspace — layered instructions, hooks, skills, and repeatable Claude Code workflows.
- openai-codex-workspace — a Codex workspace organized around
OBSERVE → IDENTIFY → PROPOSE → APPLY → TEST → RECORD. - claude-pilot-suite — benchmark-driven execution playbooks for different Claude model tiers.
- cc-workspace-docs — VitePress documentation, engineering lectures, research mirrors, and validation workflows.
- output-compress — opt-in output compression with model-tier caps and a deterministic fidelity gate.
- caveman — context and input-token reduction for agent workflows.
- CL4R1T4S and TPKI-Tri-Pattern-Knowledge — transparency research and structured knowledge patterns for AI-assisted work.
BrewMyMac · bob-plugin-openai-polisher · bob-plugin-openai-translator · Loon-scripts · Surge-scripts
| Focus | Main technologies |
|---|---|
| Apple apps | Swift, SwiftUI, AppKit, XCTest, XcodeGen |
| Input methods | Swift, C++, RIME/librime, shell |
| Local-first tools | Rust, Electron, Python, SQLite |
| Web products | React, Vite, TypeScript, JavaScript |
| Agent systems | Markdown, YAML, JSON, Shell, TypeScript |
- Local-first — keep user data close to the user whenever possible
- Privacy-respecting — treat backups and personal records as sensitive
- Reproducible — make builds, releases, and agent workflows explainable
- Evidence-backed — tests, CI output, screenshots, and release notes are part of the product
- Specification-driven — clarify product decisions and acceptance criteria before expanding implementation
- Stoat Keyboard release and regression testing across iOS versions and dictionary variants
- Unifying Linkory, Linko, and Hoppa around a reliable local-first link lifecycle
- Iterating on Luho, Minoto, and Swoop as native Swift products
- Improving LINE backup integrity, extraction guidance, and restore verification
- Making agent workspace hooks, documentation, compression, and validation easier to trust
Research → Specify → Build → Verify → Release