# ╔════════════════════════════════════════════════════════════════════╗
# ✗ Build AI models ✗ Write prompts ✗ Fine-tune LLMs
I build the missing infrastructure between AI models
and real-world tasks. The layer underneath —
execution engines: "run structured tasks, not vibes"
persistent memory: "agents that remember across sessions"
structured debugging: "diagnose, not guess"
system awareness: "know when to build, when to back off"
Background: "OS kernels in Rust · cross-platform apps · dev tools"
Focus: "AI agent infrastructure"
# ╚════════════════════════════════════════════════════════════════════╝| Project | What & Why | ||
|---|---|---|---|
 | |||
 | Nerve  | System awareness for AI agents — 22 metrics, 9-dim scoring    |  |
 | VibeBetter | AI engineering insight — measure AI effectiveness, track structural risk    | |
| HuLa | Cross-platform IM desktop app — full-stack from architecture to delivery   |  |
 | |||
 | sprint-for-agent | Task execution engine — turns instructions into verifiable steps   |  |
| know-for-agent | Knowledge compiler — persist tacit knowledge, structured docs, recall before code changes   | |
| diagnose-for-agent | Deviation diagnosis — find why results are wrong, not just fix symptoms   |  |
| decay-for-agent | Health monitor — catch project decay before it becomes tech debt   |  |
| evolution-for-agent | Self-evolution engine — agents learn from their own mistakes   | |
| skill-wrapper-for-agent | Skill quality assurance — lint instruction defects, simulate behavior chains   | |
 | |||
| MiniOperationSystem | Minimal OS kernel — memory management, interrupts, scheduling  |  |
 | FlowGen | Terminal workflow SDK — interactive CLI from JSON/YAML DSL    | |
| SmartCommit | Git commit generator — AST-level diff → Conventional Commits   | |
| VibeCodingRules | Curated rules & conventions for AI-assisted coding  | |
展开查看技术成长时间线 ▶
2018 git init life.git # 开始在 GitHub 上留下痕迹
│
├── 读源码、fork 优秀项目、建立技术品味
│
2020 rustc hello_kernel.rs # 用 Rust 写第一个 OS kernel
│
├── 理解中断、内存管理、调度器
├── 从此相信：想用好工具，先理解机器
│
2023 cargo build --release hula # 第一个独立产品：跨平台 IM
│
├── Tauri + Vue3，从架构到交付全栈负责
├── 学到：做产品和写代码是两件事
│
2025 npm publish smart-commit # 开始造开发者工具
│
├── 发现 AI agent 的真正瓶颈不是模型，是基础设施
├── 没有执行引擎、没有记忆、没有调试框架
│
2026 ./agent-infra/sprint v4.0 # All in AI agent infrastructure
│
├── sprint: 任务流水线 + 锚点验证
├── know: 跨会话知识持久化
├── diagnose: 通用偏差诊断
├── decay: 项目健康度监控
└── evolution: 元认知自我进化
"每一层都是上一层的地基。"
展开查看系统信息 ▶
tx@dev
████████████████ ──────────────
██ ██ OS: macOS (btw i use terminal)
██ ██ ██ ██ Shell: zsh + custom workflow
██ ██ Editor: Neovim
██ ██████████ ██ Theme: Clean White
██ ██ ─────────────────────────────
████████████████ Languages: Rust, TypeScript, Python,
Shell, Lua
─────────────────────────────
Interests: OS internals, AI agents,
dev tooling, photography
Motto: Build deep. Ship real.
Push forward.
─────────────────────────────
Fun fact: This profile is co-authored
with an AI agent.
展开查看随机想法 ▶
"AI 不会取代程序员，但会取代不理解 AI 的程序员。 不过，AI 也不会取代理解机器的程序员—— 因为总得有人给 AI 写基础设施。"
"最好的工具是你感觉不到它存在的工具。 最好的基础设施是你不需要思考它的基础设施。 我做的就是这种东西。"
"写 OS kernel 教会我一件事： 所有上层的优雅，都建立在下层的残酷之上。 理解残酷，才能设计优雅。"
The best tool is the one that disappears into your workflow.
If you've scrolled this far — maybe we should talk.