TX

Simplicity is the ultimate sophistication.

[Ben Vinegar @bentlegen]

hunk v0.21.0 is out - it's a big one:

- 💬 threaded + editable notes
- 🌲 file tree view + better responsive layouts
- 🔍 more git/jj diff operands
- ⚙️ extension APIs: dimmable code blocks, CLI subcommands, & more event hooks
- 🎨 color fixes for 25 themes

[Shubham Mishra 🐐 @shubh6200]

Really interesting paper on BtrLog, a reusable cloud logging service that cuts WAL append latency by up to 4× vs EBS while keeping durable archival cheap.

https://t.co/GGccfdC1f8

[Hw213 @Hw213w]

من داخل المروحية يسقط حجر مباشرة في فوهة
البركان المشتعلة.. ماذا سيحدث له عند وصوله إلى
الحمم؟

[Animation Universe 🌌 @animiuniverse]

This movie is so brutal but shown as it is for kids.

[TX @tyronecai]

@laogui 可以了，
多谢啊

[chuplung @choopyplug1]

Google's team just dropped a 9-page PDF on Harness Engineering - one formula replaced prompt engineering:

the twist: same Claude Sonnet, same benchmark - change only the harness

the formula: Agent = Model + Harness

here's the playbook in 6 steps:

step 1 → add guides - AGENTS.md, rule files, constraint docs - each line is a past agent failure turned into a permanent fix

step 2 → add sensors - linters, tests, validation scripts the agent runs on its own output before a human sees it

step 3 → build the agentic loop - plan, execute, verify, fix - bounded retries, budget caps, escalation when stuck

step 4 → externalize memory - the model forgets every session. the harness remembers state, decisions, and artifacts across all of them

step 5 → enforce permissions - which tools, how many writes, what needs approval. the model doesn't enforce safety. the harness does

step 6 → wire observability - track every tool call, cost, retry. trip wires fire when behavior drifts

the result: your agent stops being a demo and starts being infrastructure - every failure makes the system permanently better, not just the next conversation

this 9-page PDF is what comes after prompt engineering and context engineering

bookmark and build the full system from the article below ↓

[Chasen @chasen_liao]

我实测发现 。。。很多模型会默认调用 bash 命令，而不用 Pi 内置的 grep、find 指令，直接使用 ls、cat 这些命令行工具，应该是 模型没有针对 harness 调优过，所以很多默认会直接使用 bash，而直接使用这些会导致上下文膨胀的很厉害

bash 里的 grep 是上下文杀手。它不认识 .gitignore，递归一跑，node_modules 里每一行匹配都原样灌进 Context，token 哗哗地掉。内置搜索至少有 limit、分页和排序，输出是可控的，但是模型会不喜欢用。。。

更有说服力的是，帮我把搜索纪律写进 AGENTS.md 的那个 agent，自己就在 bash 里跑了 grep 查 pi-fff 源码，被我当场抓包。惯性走 bash 不怪模型，是因为很多模型训练语料就是这样，没有专门为 Pi 调优过

所以我重新加强了全局规则，现在 AGENTS.md 的工具节就两行：

1️⃣搜索用 grep/find 内置工具，多 OR 词用一��� multi_grep，必须走 bash 时用 rg
2️⃣定位后用 read 的 offset/limit 只读命中附近，工作区外已知文件直接 read

为什么留 rg 这个口子呢？rg 默认尊重 .gitignore，跳过二进制和隐藏文件，真要用 bash 搜，输出经常比 grep 干净一个数量级

这样子你的上下文就干净了

[War Times @wartirnes]

Witness the devastating power of the BLU-96/B thermobaric bomb.

This 2,000-pound weapon disperses fuel into the air before ignition, using the surrounding atmosphere to produce a massive high-pressure blast that can reach into enclosed spaces.

[Deadly Weapons @deadlyweapns]

Insane New A7 M109 Paladin howitzer.

[kabikabi @jakevin7]

Pi 的新架构 和 Maka 告诉我们 harness 的答案早就在数据库论文里了!!

Maka的核心作者很多都是 database 背景出身的。
当Pi 的 harness v2 文档出来后，Maka内部群大家直接愣了，因为它和我们在 Maka 里做的架构几乎一致：

持久化事实 → 派生运行状态 → 派生模型上下文 → 派生 UI → 崩溃后重新归约。
连 tool 的边界都一样：先写执行意图，再做副作用，再写结果：
Pi: tool_started intent → execute → tool-result entry
Maka: T1 toolDispatch → execute → T2 function_response

Pi V2 和 Maka从各自的路径出发，最后设计出几乎一样的架构,已经说明了很多问题了。
我已经感觉到 Agent harness 这一层正在形成一些共识。
这个共识其实就是 OS 和 db 那一套——预写日志、事件溯源、可恢复的持久化操作。
所以��程没有银弹，当我们最后做了很多工作之后，发现核心原理还是软件工程的那一些内容。

https://t.co/oU5BK2AfcQ

[Deadly Weapons @deadlyweapns]

The Finger of Death artillery shell!

[孙悟元 @wuyuandev]

冷知识：博物馆牌子上的距今时间，是距离1950年，而不是今年。

[瓜哥玩AI @MatrixCoreX]

大清时期，中医们已经拥有了几千年来积攒的临床经验了。

中医们发现，如果人大量的摄入铅，会让人昏昏欲睡��精打采。

于是中医们将铅制成铅白霜，并作为安神汤的重要药材。

这方子虽然坑，但是很少有人用，老百姓吃饭都是问题，何谈闲的没事来副安神汤呢那，有没有一个群体，又有钱，又活力四射，需要经常服用安神汤呢。

有的，那就是大清的皇子和公主们。

由于不希望自己的孩子过于有活力，在宫内惹是生非，后宫的妃嫔们经常会要求自己的子嗣服用安神汤。皇子公主喝完之后，立刻铅中毒，自然就安神了。

[Chad Infi𓄿 @chad_infi]

It all started with a male ego

[Deadly Weapons @deadlyweapns]

cold launch" technique—one of the most fascinating engineering solutions in the defense industry.

[Airforce Geek @PrecisionLabs0]

This wasn’t science fiction. It was Operation Credible Sport. 🇺🇸

In 1980, the XFC-130H was turned into one of the wildest aircraft ever tested packing 30 rocket motors to make a heavily modified C-130 capable of extreme short-field operations.

It had MK-56 boosters for takeoff, Shrike rockets for descent and ASROC rockets for braking
The plan was to land inside Tehran’s Amjadieh Stadium, extract the American hostages, and get out.
The footage shows the insane part: rockets firing as the Hercules accelerates like a fighter jet, surrounded by smoke and flames.

The mission was ultimately canceled but the engineering experiment was REAL.

[23 DERECE @yirmiucderece]

Yeni mucize bulundu!

"14 günde düz göbek, günde sadece 2 dakika."

Sosyal medyada dolaşıma giren bir videoda spor eğitmeni, "14 günde düz göbek" vaadiyle dikkat çekti.

İddialı formül, "Spor salonlarını emekliye mi ayıracak?" yorumlarını da beraberinde getirdi..

[Deadly Weapons @deadlyweapns]

How Fast Does the Swedish Navy's Stealth Bofors Naval Gun Fire?

[Radio @video_00]

💢 "معركة شمال روسيا "

في غابات سيبيريا شمال روسيا تم نقل عدد من النمور البنغالية الملكية لتتعايش مع المنطقة.
وأثناء ترقب النمر لفريسته لمح حيوان الائيل الضخم وقرر ان يكون وجبته الدسمة لمدة أيام

ولك�� بعد ان بداء النمر بالافتراس ظهر له دب ضخم🐻
الدب كان منزعجاً بقوة كيف ان النمر يدخل محميته ويصطاد فرائسه!!

فبدات معركة شرسة بينهم أنتهت بضربات قوية من مخالب الدب على النمر مما جعل النمر يفقد قوته وينسحب من المكان وهو جائع.