4 个帖子 - 3 位参与者 阅读完整话题
codex新版tui的context显示太傻了吧... 搞七捻三 [image] ？？？？？ 随手升了个级 之前好好的百分比 怎么换成了进度条了。。。从一眼看变成要预估了。。。 [image] 原来已经有人反馈了啊 [image] 好的 说的是根据用户反馈改的。。 然后被15个 是何意味 前几天刚提这个，那时候也有不少issue说这个问题。 今天更新到0.121.0 百分比居然回来了 不过… 这又是什么新活？ 8 个帖子 - 6 位参与者 阅读完整话题
虽然目前有一个 zenobi-us/pi-dcp ，但是已经两个月没更新了。 之前用opencode的时候感觉 Opencode-DCP/opencode-dynamic-context-pruning 很好用，可以不声不响地把context 保持在一个很低的水平。 换到pi-agent后，context动不动就飙升到40-50%。试了一下 mksglu/context-mode 和 MasuRii/pi-rtk-optimizer ，效果都不明显。用pi-agent的佬们，有没有什么好的管理办法？ 1 个帖子 - 1 位参与者 阅读完整话题
求一个有text-embedding-3-small或者Qwen3-Embedding-8B的公益站，感谢佬们 3 个帖子 - 2 位参与者 阅读完整话题
devnexus is an open-source cli that gives agents persistent shared memory across repos, sessions, and engineers. It maps out dependencies and relations at the function level, builds a code graph, and writes it into a shared Obsidian vault that every agent reads before writing code. Past decisions are also linked directly to the code they touched, so no one goes down the same dead end twice. Still building it out but I would love to hear any thoughts/feedback Comments URL: https://news.ycombinator.com/item?id=47812829 Points: 4 # Comments: 0
I've been presenting at local meetups about Context Engineering, RAG, Skills, etc.. I even have a vbrownbag coming up on LinkedIn about this topic so I figured I would make a basic example that uses bedrock so I can use it in my talks or vbrownbags. Hopefully it's useful. Comments URL: https://news.ycombinator.com/item?id=47808956 Points: 3 # Comments: 0
Lately I've been working on a small tool to generate realistic handwriting animation from any text. Any feedback is welcome, hope you enjoy :) Comments URL: https://news.ycombinator.com/item?id=47805078 Points: 2 # Comments: 1
2019年初 GPT2还被媒体称为文本生成器（text generator），Anthropic现CEO Dario Amodei还在Openai的时候； 7年后… 2026年初 Claude Mythos 还被称为AI LLM的时候。 7年后… 2033年… 2019 Better language models and their implications | OpenAI Reddit - New text generator built by OpenAI considered too dangerous to release OpenAI built a text generator so good, it’s considered too dangerous to release 2026 X.com - We do not plan to make Mythos Preview generally available… 4 个帖子 - 4 位参与者 阅读完整话题
目前尝试组合以下几个工具： Augment Context Engine (ACE) MCP：语义检索减少全量文件投喂； JuliusBrussee/caveman Skills：减少冗余对话； rtk-ai/rtk CLI Proxy：过滤终端输出。 体感能省下至少 50% 的 Input Token，输出质量没有明显降低，想请教下还有哪些方案。 3 个帖子 - 3 位参与者 阅读完整话题
claude ▐▛███▜▌ Claude Code v2.1.112 ▝▜█████▛▘ Opus 4.7 (1M context) with xhigh effort · API Usage Billing ▘▘ ▝▝ C:\Users\dingy ❯ 你好 ⎿ 503 {“error”:{“code”:“model_not_found”,“message”:“No available channel for model claude-opus-4-7 under group auto (distributor) (request id: 202604170325133370552998268d9d6wLWX86kr)”,“type”:“new_api_error”}} Retrying in 12s · attempt 10/10 Nebulizing… (2m 53s) ⎿ Tip: Use /btw to ask a quick side question without interrupting Claude’s current work 这种情况是不是claude官方做了什么限制？ 现在国内中转站完全用不了了 3 个帖子 - 3 位参与者 阅读完整话题
Claude Opus 4.7 = Opus 4.6 + GPT 5.4 - (1m - Gemini 3.1 Pro Context) - Role Play ps:仅根据目前了解 [1] 作调侃,并没有实际上手体验测试过 1. opus 4.7 （high）cursor里面的，糖果问题和红绿色盲，锐评一下 - 开发调优 - LINUX DO 2. 来自官方System Card的Claude Opus 4.7牙膏倒吸实证 - 搞七捻三 - LINUX DO 3. 怎么感觉Opus 4.7 语气有点像GPT - 搞七捻三 - LINUX DO 4. claude4.7opus发布了,我生成了两张图可以概括4.7opus能力 - 开发调优 - LINUX DO ↩︎ 4 个帖子 - 3 位参与者 阅读完整话题
What's New in v0.3 Retry failed transcriptions – history entries that failed can now be retried with saved audio Cancel with ESC – press ESC during recording to cancel without creating invalid entries More stable long recordings – fixed truncation issues for extended sessions Fn key support – custom hotkeys now support Fn key combinations Comments URL: https://news.ycombinator.com/item?id=47795617 Points: 2 # Comments: 1
Article URL: https://zxclip.com/ Comments URL: https://news.ycombinator.com/item?id=47795251 Points: 2 # Comments: 0
Article URL: https://ilha.build/ Comments URL: https://news.ycombinator.com/item?id=47791875 Points: 3 # Comments: 0
看了很多harness的文章和实现 我大致总结了一下， hook就是将agent的生命周期拆分成几个钩子（简单示例） before agent before model input before tool exec after tool exec after model output after agent 在不同的时期程序介入处理 然后将hook实现+工具+skill+agent上下文打包成插件 就能实现一个基本的harness系统的架子 包括记忆、状态、鲁棒性、观测性、权限约束、上下文压缩 都能以hook+工具+skill+agent上下文打包一个插件来实现 并且很容易通过移除、添加不同的类似插件来实现扩展性 6 个帖子 - 4 位参与者 阅读完整话题
Article URL: https://atticsecurity.com/en/blog/why-llms-hate-fake-data-token-proxy/ Comments URL: https://news.ycombinator.com/item?id=47778087 Points: 2 # Comments: 2
现在的compaction感觉不是很靠谱，无论是丢信息啊还是之类的。刚刚突然在想，之前经常因为一些原因一个session烂掉，这个时候我想续上这个session，我就会直接复制这个session的最后大约20k（我也不太清楚）的内容，然后作为prompt塞进下一个session就好了，然后对话就续上了 感觉比compaction便宜还快 不如直接做成每次满了，就去掉前80%然后就好了 虽然最好的方式显然是滑动窗口，但是由于缓存便宜，所以完全行不通。 评价是： context rot的根源是大模型太贵了（用缓存才稍微正常一点）（大模型厂商怎么这么坏，就知道节省计算资源 ）（虽然是缓存计算成本低是客观事实）（虽然缓存确实对用户的钱包更友好一点） 1 个帖子 - 1 位参与者 阅读完整话题
Article URL: https://github.com/manojmallick/sigmap Comments URL: https://news.ycombinator.com/item?id=47776309 Points: 3 # Comments: 6
I found that I often shy away from posts on Hacker News or sites like ScienceMagazine because I don’t understand every second word on the page. A simple dictionary extension doesn’t really help in these situations, because I’m not looking for the literal meaning of a word. Like one time I was reading a post about Rust compilers and the dictionary extension just failed hilariously.. “Rust” (the programming language, you fool!! not the thing that forms on iron when it oxidises -__-). What I actually need in these moments is context-aware explanations of concepts, not dictionary definitions. AI turns out to be pretty good at that. So I built a very simple tool for this — under ~300 lines of code — that has ended up saving me a ton of time. It works pretty simply: it takes the highlighted text, parses the HTML of the page for surrounding context, and sends both the page content + selection to an LLM provider of your choice (OpenAI, Gemini, or Claude). You can also reorder which provider you prefer. That’s basically it. I don't collect any data or anything. Here’s the GitHub link: https://github.com/oishika10/rhino I’m pretty sure this can be improved a lot, so I’d really appreciate feedback or ideas. This is the first project I’m sharing publicly like this. I enjoyed building this because made me realize how small, simple tools can still be surprisingly powerful and save me a ton of time in life. Comments URL: https://news.ycombinator.com/item?id=47776069 Points: 1 # Comments: 0
ClaudeCode中配合会对复制的长文自动折叠, 格式如: [Pasted text #1 +3 lines] 请教各位佬有解决办法吗? 看到官方好像有不少Issue提到这个问题, 看起来是官方暂时不支持这个折叠的开关吗? 对于日常使用Typeless之类的语音输入不太友好~ 1 个帖子 - 1 位参与者 阅读完整话题