要件定義が苦手な人はgrill-meを使えgrill-meを使えgrill-meを使えgrill-meを使え頼むからgrill-meを使ってくれ ※AIと人間で共通理解が得られるまでこちらに徹底的に質問を行うSkillです。使い方はこちら↓ https://ai.itokoba.com/archives/861/#2-grill-me

总结了一下评论区， 提防 GPT-5.6 Sol 过度工程化 一句话概括的话就是： 规划可高，执行要轻，约束先行。 - 先复述意图、范围、非目标、验收标准 - High 只做规划，Low / Luna / Terra 落地 - 禁止全程 Ultra / max，禁止重 skill，禁止多 Agent 乱开 - 不可逆操作等确认；Git 回滚不算不可逆 - 完成前检查：测试过、diff 小、无多余抽象 不能证明必要的设计，默认不做。 同时，我让 Grok 4.6基于评论区里的高置信度内容设计了一个Agents.md👇不过我认为可以先测试，再根据实际情况看要不要再跟AI交互已简化这个方案。因为grok依然认可测试的重要性，只是压住无效以及与本轮无效的测试。

深有同感：让 Sol 审一段代码 它总能找出几个问题。 不管这段代码有多大 也不管你已经让它审过多少轮 按它的意见改完了 再审一遍 它还能再找出几个新问题， bugs 是无限的。 大家可以试试看：新开上下文再审，第一轮修完的问题，第二轮就是有新的出来，现在 AI code review 最大的问题是，假如你不断的审下去你就会发现后面的问题其实都是硬找的问题，把风格偏好说成缺陷等等。 所以在 PR review 上我还是一直在用我的 https://github.com/vikingmute/review-forge，多 agents 审查，汇总通用问题，手动 check 要修复的问题，fix 和 verify 。

用以下Agents.md可以缓解GPT-5.6 Sol的过度设计和测试。我在做一个更细致但更重的方案，做好之后发出来。 ## Design simplicity (anti-over-engineering) Over-engineering is the most frequent correction in this workspace. These are hard defaults and they apply INSIDE the requested scope, not just to adjacent changes: - No backward compatibility, migration shims, legacy fallbacks, or old-data backfill unless the user explicitly asks for them. Assume breaking changes are acceptable by default. - Words like 完美 / 长远 / 通用 / 一定不能影响 / "perfect" / "l…

ただの酸素系漂白剤なのにクソ高えオキシクリーン使ってる奴らは情弱

His child accidentally released the handbrake…He chose love over anger. 🥹❤️

靠…..看著都一陣哆嗦🫨

@Sittata_Sub こんなんあるで〜。fedoraなら動作確認できた。

New Crop Circle Poppy Field Wiltshire, England June 24, 2026 YouTube: Stonehenge Dronescapes

🚨 Iran closes all diplomatic and indirect channels of contact with the US, according to Tehran Times

do u think im cute?

This is my quant

The guy who invented cheese is a hero 🧀

Ibiza 2000 vs Ibiza 2024

btw once you have all these security in place this is what it unlocks: https://x.com/elvissun/status/2025920521871716562?s=20

Qwen just dropped the open-source Claude Code killer. It's called Qwen3-Coder-Next and it's genuinely wild. Only 3B active parameters but beats models with 10x-20x MORE parameters on SWE-Bench-Pro. PLUS: They launched Qwen Code CLI, the best open-source alternative to Claude Code. 1,000 free requests/day. This is the model that proves size is a lie Here's everything you need to know

I’ve legit watched this synth battle 13 times on repeat.

How I be eatin the pussy

I poured my 10 years of teaching experience into a skill. It's called /teach, and it can teach you anything. Here's how it taught me to solve a Rubik's cube:

this is genuinely and comically insane he’s like playing in a league of his own the rest of us are closer to hunter gatherers than to this man

When philanthropy looks more like a firmware update. Gates, Epstein, and the cloud you can't log out of.

🚨 Per the Washington Post: The US military is preparing boots on the ground in Iran, a 2-month operation. Everything changes from here.

estoy tan decepcionada de ti, pero aún recuerdo cuando llegaba feliz a mi casa pensando que había encontrado a alguien diferente e incapaz de lastimarme

🚨 Breaking: Claude can now create a presentation in 5 minutes. COPY AND PASTE these PROMPTS and watch the magic happen.

skills are the new apis so many new companies are just going to be skills

The most important skill to learn in the next 10 years

Black Sabbath - 'War Pigs' (pigeon version) 🕊️ Pigeons are government drones 🕊️ Iran. Israel. Trump. The military-industrial complex in full force again.

@0xBoomz @MetaDAOProject $MOCK is one step ahead 👀

@JamesWynnReal I’m long $MOCK 💥

Victoria Beckham’s son, Cruz Beckham, has dropped his debut single “Optics” which reached #79 on the UK Singles Chart.