DQ10

take it easy

[DQ10 @le0l11]

@geekbb 黄Sir 也是benchmark maxing

[DQ10 @le0l11]

@ivanalog_com 那就不知道了，以前cline kilo那些编程工具针对一些tool call不行的模型就这么干的，感觉还是训练问题

[DQ10 @le0l11]

有意思，coding lite用户没有3.6plus 其他的却可以免费调

[Qwen @Alibaba_Qwen]

🚀 Qwen Code v0.14.0 – v0.14.2 are now available

Channels：Control Qwen Code remotely from Telegram, DingTalk, or WeChat — send a message from your phone, get results on your server
Cron Jobs ：Schedule recurring AI tasks — auto-run tests every 30 min, pull & build every morning, monitor logs on a timer
Qwen3.6-Plus ：New flagship model with 1M token context, 1,000 free daily requests
Sub-agent Model Selection：Assign different models to sub-agents — use a powerful model for the main task, a fast one for subtasks, save tokens without sacrificing quality
/plan：Enter planning mode before execution — AI maps out all files and steps first, you confirm, then it executes
Follow-up Suggestions：AI suggests 2-3 next steps after completing a task — "Add unit tests?" "Check similar files?" — click to continue
Adaptive Output Tokens ：Default 8K output, auto-escalates to 64K when truncated — no more manually tuning max_tokens
Ctrl+O Verbosity Toggle ：Switch between verbose and compact output mid-conversation — debug mode when you need it, clean mode when you don't

📋 Full changelog: https://t.co/zPMyCVIskN

[DQ10 @le0l11]

通过上层做的方式，会比hook机制更强制。 在流程节点强制执行 而不是简单的在hook里面弹一句prompt 让ai注意流程，这也就是为什么superpowers需要使用hooks, 纯skills的编排很难做到工作流的闭环

[DQ10 @le0l11]

都在说harness 我想问个问题 用claude code接国产模型 比如说kimi glm的时候

agents.md 要求启用subagent分别走 excute -> evaluate

经常漏掉evaluate步骤，有啥办法能约束？

[DQ10 @le0l11]

目前发现最简单的解法是 通过上层直接调用claude code来解决，通过tmux send keys或者 -p
来处理，强制拟定工作完成后必须要触发子agent做校验

[ensue @ensue_ai]

Coordination scales intelligence.

Inspired by @cursor_ai’s work on large-scale autonomous coding, we show how multi-agent swarms with shared memory can solve complex reasoning problems (e.g. Putnam-level math) - reproducible with Claude Code.

[Dan McAteer @daniel_mac8]

Harness engineering is as important as model capability scaling.

AI Agents are 50% a harness story.

"Natural-Language Agent Harnesses" proposes moving harness logic out of code and into the native language of LLMs:

Natural Language.

Turns the agent harness into a portable artifact that you can experiment with, improve and execute in a shared runtime.

Agent harnesses will mature into a first class citizen of the AI Agent ecosystem.

We will have people who are Agent Harness engineers and experts.

[DQ10 @le0l11]

@NPC888666 本质就是没啥兴趣爱好 朋友都是建立在相同的消费观上的，男的都是低物欲多 自然与外界没啥链接

[DQ10 @le0l11]

实际使用也发现了cursor寻找代码的速度很快，底下也有一些开源实现 回头接到pi agent里面试试

[Cursor @cursor_ai]

Cursor can now search millions of files and find results in milliseconds.

This dramatically speeds up how fast agents complete tasks.

We're sharing how we built Instant Grep, including the algorithms and tradeoffs behind the design.

[DQ10 @le0l11]

大模型开始进入了一种 你爷爷一开源我爷爷就原创的奇妙阶段

观日本乐天 composer2有感

[Augment Code @augmentcode]

https://t.co/Af9pI4L8ML

[DQ10 @le0l11]

@shaoruu @evanyou the cli doesn't support plugins and rewind，when will we have it

[DQ10 @le0l11]

@benvargas I've used it, but don't find anything special and when using openai compatible, i can't set the context window length and etc..

[DQ10 @le0l11]

没人吐槽长桥的新图标？ 看上去像屏幕坏了 #longbridge

[DQ10 @le0l11]

glm5涨价，说明目前token成本虽然在飞速下降，但是新模型 好模型的成本依然很高 尤其是处理复杂任务时需要更好的模型 更多的token。

也侧面说明了国内算力确实是缺的，过剩之说还很长远 ，可以再看看找找国内的标的

至于AI到人的商业闭环，现阶段不用过多考虑了

[DQ10 @le0l11]

@LZRationalnvest 这么说 还得加点

[DQ10 @le0l11]

vibe了一个 memory+pattern 插件

多轮对话后或者session结束时，利用llm对对话进行提取，评估出经验模式，同时要求每次任务前必须从过往经验模式中进行查询 先用一段时间看看

这部分感觉有点像https://t.co/IJArvAfvxq该做的事，不过能够自动沉淀经验，做到self-evolving应该更好

[DQ10 @le0l11]

@yetone 也需要模型能力支撑，有些模型就是死活不调skills 改成定义tools就好了