
从整人专家到 AI 修改器:三十年,一个循环
风灵月影要收费了,网上的修改器到处是病毒。作为一个从 1996 年就开始用整人专家 FPE 的老玩家,我决定自己动手,用 Claude + Python 造了一个通用型游戏修改器。三十年后,修改游戏的原理没变,变的是谁在帮你写代码。
Sharing insights on learning, data-driven innovation, AI thinking and my life

风灵月影要收费了,网上的修改器到处是病毒。作为一个从 1996 年就开始用整人专家 FPE 的老玩家,我决定自己动手,用 Claude + Python 造了一个通用型游戏修改器。三十年后,修改游戏的原理没变,变的是谁在帮你写代码。

用了三年的 Otter AI 越来越抠门,从 6000 分钟缩水到 1200。作为一个非 Native Speaker,我决定自己动手,用 Deepgram + GPT-4.1 Mini 花一周时间造了一个更好用的会议转录工具。这件事让我感慨:在这个时代,创造力真的可以被无限放大。

I added a Telegram bridge to my CodePilot project and realized it's becoming OpenClaw. Opus 4.6 did most of the work. When building is this easy, what's actually worth building?

I spent weeks building a web-based Claude Code interface for remote access from my phone. Then Anthropic shipped a native remote control feature. My version is still better — but the clock is ticking.

Google Antigravity IDE 被用户通过 OAuth 漏洞薅算力羊毛,导致大规模封号。从技术原理到行业对比,一个关于"信任"与"架构"的故事。

AI coding assistants forget context between sessions, reintroduce fixed bugs, and make baffling 'improvements.' Tests are the guardrail that survives the amnesia. Here's what I learned building a real project with Claude Code — and what Kent Beck, Martin Fowler, and the data say about testing in the age of AI.

The Agent Skills spec is open. The example skills are open. But what about the runtime that loads, selects, and orchestrates them? Every major tool — Claude Code, Codex, Copilot, OpenClaw — built their own. Here's what's open, what's not, and what you actually need to build.

Traditional workflow builders like n8n and Zapier connect apps with rigid If-This-Then-That logic. But a new paradigm is emerging: composable AI skills orchestrated by intelligent agents. Here's what's actually working, what's not, and why the shift matters.
Leave a message, say hello!