I build AI-agent operating systems for markets, content, and product creation.
用 AI Agent 把交易、内容与产品创造，变成可以持续运行和学习的系统。
上海 · Builder · Systems thinker
| Entry point | Registry | What it contains |
|---|---|---|
| Trading | Public | Market data, research, strategies, engines, interfaces and full systems |
| Content | Public | Discovery, production, publishing and growth |
| Product Lab | Public | Mature products, one repo per row |
| Agent | Public | Runtimes, orchestration, skills, memory and tooling |
| Park OS | Private | Canonical universe snapshot, governance and lock authority |
The registries contain the full catalog. This profile keeps only the map and a small amount of system context below.
我不是在收集互不相干的 repo。我在搭三套相互连接的操作系统：
| System | Outcome | Current focus |
|---|---|---|
| Trading OS | 从市场事实走到受约束的交易决策与复盘 | Paper-first 闭环与可审计证据 |
| Content OS | 从内容信号走到跨平台成品与反馈学习 | 独立 capability 串成生产系统 |
| Product Lab | 从成熟产品链接到真实使用证据 | 一 repo 一行，直接进入产品入口 |
| Agent | 从 runtime 到 skills、memory 和 orchestration | 通用 Agent 基础设施 |
| Park OS | 维护 Park 自己的身份、治理和 universe 真源 | canonical snapshot + lock authority |
READY 可用入口与核心结果明确 · BUILDING 正在形成完整产品 · EXPLORING 仍在验证方向
Outcome: 把多市场数据转成有来源、有风控、有执行证据的决策闭环。当前坚持 paper-first；不在这里宣称真钱能力已经开放。
Market facts Understanding Decision Learning
01 Data ──▶ 02 Intelligence ──▶ 03 Signal ──▶ 04 Method routing
READY READY BUILDING BUILDING
│
▼
05 Backtest ──▶ 06 Risk plan ──▶ 07 Paper execution ──▶ 08 Journal
READY BUILDING BUILDING EXPLORING
| Product | Role in the system | Status |
|---|---|---|
| datafeed | ticker + timeframe → multi-market OHLCV | READY |
| intel | 10+ sources → scored, clustered market events | READY |
| equity-research | evidence snapshots → A-share investment-committee research | BUILDING |
| backtest | strategy definition → win rate, payoff and drawdown | READY |
| trading-strategy | Canonical DCA/Grid strategy → deterministic plans, previews and replays | READY |
| standard-kline | OHLCV + provenance → trustworthy chart surface | READY |
- Now — 强化从行情、情报、策略到 Paper 执行的可审计闭环；把“代码存在”和“真实运行证据”分开。
- Next — 用完整周期证据验证多资产闭环，再决定哪些能力值得进入更高风险阶段。
More context
- quant-data-pipeline 是早期多市场数据与信号能力的集成底座。
- 私有执行与风控实现不会从 Profile 暴露；公开页面只描述能力边界和已验证状态。
- Backtest 是辅助证据，不自动等于策略可交易，更不等于 live-ready。
Outcome: 让创作者负责判断与表达，让系统处理发现、获取、理解、生产、组装、分发和反馈。
Discover Understand Create Learn
01 Signals ──▶ 02 Acquire ──▶ 03 Extract ──▶ 04 Curate
BUILDING READY READY EXPLORING
│
▼
05 Rewrite ──▶ 06 Assemble ──▶ 07 Publish ──▶ 08 Performance
READY READY BUILDING EXPLORING
| Product | Role in the system | Status |
|---|---|---|
| content-intelligence | social data → trends, patterns and topic signals | BUILDING |
| content-downloader | platform URL → normalized media + metadata | READY |
| content-extractor | video / image / article → structured text | READY |
| content-rewriter | source material → platform-specific drafts | READY |
| videocut | talking-head footage → edited video assets | READY |
| daily-newsletter | source feeds → selected Chinese daily brief + receipts | READY |
- Now — 独立能力已经覆盖获取、理解、改写和视频组装；重点是让它们以清晰合同协作，而不是继续堆工具。
- Next — 补齐 curator 与 performance feedback，让选题质量和发布结果能够回流到下一轮生产。
More context
- seedance-expert 把视频创意转成可执行的多模态生成提示。
- AI-videos 探索虚拟人物换装与动作迁移工作流。
- 已归档的 orchestrator 和 workbench 保留为历史证据，不再占据主地图。
Outcome: 把独立产品想法做成 ready for use 的产品，再用真实任务验证价值，把缺口送回 Build。
Intent Build Gate Use
01 Explore ──▶ 02 Product build ──▶ 03 Readiness ──▶ 04 Real tasks
READY BUILDING READY BUILDING
│
▼
06 Improve ◀── 05 Evidence
READY READY
| Product | Role in the system | Status |
|---|---|---|
| proactive-explorer | existing product → evidence-backed next direction | READY |
| doc-driven-dev-workflow | intent → reviewable development stages and guards | READY |
| wechat-miniprogram-shipping | Mini Program intent + receipts → evidence-gated shipping path, independent QA and recoverable release handoff | READY |
| repo-evals | product claims → reproducible verdict dossier | READY |
| loop | repo + contract → value-ranked issues, PRs and digest | READY |
| codex-harness | local agent sessions → project and token evidence | BUILDING |
- Now — Product Lab 已明确分成 Build 与 Use：Build 对 readiness 负责，Use 对真实任务与使用证据负责。
- Next — 把 ready handoff、真实使用、缺口复现和回流需求做成跨产品可复用的证据链。
Operating rule
Intent → Issue contract → Build → Readiness gate → Real use → Evidence → Next issue
绿测试证明代码通过了测试，不自动证明产品 ready；部署成功也不自动证明用户结果已经发生。
┌─────────────────────────┐
│ Agent Product Lab │
│ builds + tests + learns│
└────────────┬────────────┘
│
product capabilities
│
┌────────────────┴────────────────┐
▼ ▼
┌────────────┐ ┌────────────┐
│ Trading OS │ │ Content OS │
│ decisions │ │ production │
└─────┬──────┘ └─────┬──────┘
└──────────── evidence ───────────┘
│
▼
better next build
三套系统共享同一条原则：先定义结果，再建立可复验合同；系统必须诚实表达 READY、BUILDING 和 EXPLORING。
如果你也在构建 agent-native 产品、研究系统或内容基础设施，可以从相关 repo 的 issue 开始交流。最好的合作入口不是“聊一个大想法”，而是一个清楚的问题、输入、期望输出和失败边界。
Build the system. Use the system. Keep the evidence.