Jason
301 posts
Jason
@JasonBoy7
Web ✨NodeJS developer; 🤖 Agent Harness; Being an indie hacker
เข้าร่วม Temmuz 2012
151 กำลังติดตาม162 ผู้ติดตาม
开发者 Can Vardar 发现，Claude Code 里如果关闭遥测（telemetry，即向 Anthropic 回传使用数据），提示缓存时间会从 1 小时骤降到 5 分钟，他算了笔账说这相当于隐私换 12 倍性能，给 Anthropic 扣了个邪恶公司的帽子。
这条推文传开后，Anthropic 工程师 Boris Cherny 出来做了详细回应。
Claude Code 的缓存策略一直是个黑盒子，Boris 的这条推文把里面的细节讲的比较清楚了，推荐可以仔细看看。
他先澄清了一个误解：1 小时缓存并不是无条件更好。缓存写入成本更高、读取成本更低，划不划算取决于你怎么用。如果你只是跑了一次查询就走了，1 小时缓存反而浪费钱，因为你付了写入的高价却没享受到反复读取的便宜价。
实际上 Anthropic 一直在根据使用场景做精细化调整。比如子任务（subagent）很少被恢复，给它 1 小时缓存纯属白花钱，所以这类查询就保留 5 分钟。API 用户目前也没有默认开启 1 小时缓存，还在测试阶段。
关掉遥测导致缓存变短，Boris 说这其实是个连带效应：遥测关闭后，客户端的实验开关也跟着失效了，系统读到的就是默认值 5 分钟。换句话说，这不是故意惩罚，是技术实现上的耦合问题。
Boris 还透露了后续计划：很快会把部分查询的客户端默认值改成 1 小时，同时提供环境变量让用户自己强制切换 1 小时或 5 分钟。
至于12 倍性能差距，Boris 说远没有那么夸张，实际节省的 Token 量并不大。
Boris Cherny@bcherny
👋 1h prompt cache is nuanced actually. It costs more for cache writes, and less for cache reads. Whether you benefit from cheaper cache reads depends on your usage pattern -- context window size, whether the query is the main agent or subagent, etc. We have been testing a number of heuristics to give subscribers better prompt cache hit rates, which means lower token usage and lower latency, when it works. But this effect is far from uniform due to the nuance above. Say you use 1h cache for an agent, but only used the agent to make a single query -- in this case the 1h cache would be wasted and you'd be overcharged. At this point we have rolled out 1h prompt cache by default in a number of places for subscribers to optimize cache duration based on real usage patterns, but we actually keep it at 5m for many queries also (eg. subagents, which are rarely resumed so you'd be paying for them even though they do not benefit from 1h). We also are not defaulting API customers to 1h yet -- this needs more testing to make sure it's a net improvement on average. Separately, when we do this kind of experimentation, we use experiment gates that are cached client-side. When you turn off telemetry we also disable experiment gates -- we do not call home when telemetry is off -- so Claude reads the default value, which is 5m. We will soon be changing the client side default to 1h for a few queries, since we now feel good that it is a small token savings on average for those queries. We will also give you env vars to force 1h and 5m. In any case, the token savings is nowhere near 12x unfortunately. It is a small win though, that we have been in the process of rolling out to everyone. Hope the explanation helps. More here: #pricing" target="_blank" rel="nofollow noopener">platform.claude.com/docs/en/build-…
中文
We've been working on this for a while. Can't wait to hear what you think
Claude@claudeai
We've redesigned Claude Code on desktop. You can now run multiple Claude sessions side by side from one window, with a new sidebar to manage them all.
English
在 #ClaudeCode 中，使用“代理团队”（Agent Teams）时，不同模式之间的一个重要区别在于对 MCP 服务器的处理方式：
• 🔂进程内模式 (In-process mode)：MCP 实例在主代理和团队成员之间是共享的。
• 👥Tmux 模式 (Tmux mode/Split mode)：每个团队成员都拥有完全独立的 Claude Code 会话，因此各自拥有独立的 MCP 实例，互不干扰。
以使用 @playwrightweb MCP 进行自动化测试为例：
在进程内模式下，由于所有团队成员都在同一个浏览器会话中操作，会产生冲突。
解决方案：
• 如果适用，切换到 Tmux 模式（分屏模式）。
• 直接通过命令行使用 playwright-cli 或 agent-browser 来实现会话隔离。
code.claude.com/docs/en/agent-…
中文
One difference between using different modes with Agent Teams in #ClaudeCode is that when you use a MCP server, the mcp instance is shared between main agent and teammates in in-process mode, while in tmux mode, it's fully independent claude code session so that each teammate has its own mcp instance without interfering with each other.
One example is that when you use @playwrightweb mcp to do autonomous testing: when in in-process mode, the teammates will conflict since they all operate on a single browser session, 2 solutions:
- switch to tmux mode(split mode) if appropriate
- use #playwright-cli or #agentbrowser to isolate sessions from cli directly
code.claude.com/docs/en/agent-…
English
Cursor Agent 3.0 被曝“套壳” Claude Code，官方回应称为小范围测试
Cursor 的新版 Agent 被曝出“套壳”了 Anthropic 的底层技术。开发者逆向分析发现，Cursor 3.0 版本在本地运行了一个替换引擎，简单粗暴地把系统提示词和消息里的“Claude”字眼全部替换成了“Cursor”。其安装包里直接内置了 Anthropic 官方的 Agent 开发工具，并接入了一个专门为 Cursor 微调的 Claude 3.7 模型。
面对社区的质疑，Cursor 团队成员 Michael Truell 迅速给出了澄清。他表示这只是一次常规的 A/B 性能测试，目的是对比评估 Claude 官方框架和 Cursor 默认框架的优劣。这次测试非常隐蔽，只覆盖了不到 1% 的流量。
Michael 说的应该是事实，只是这种形式爆出来还是有点尴尬 😅
Jason Kneen@jasonkneen
Delved into Cursor 3.0 -- turns out there's some interesting shenanigans going on.... "The most newsworthy finding is that "Cursor Agent" is a rebranded Claude Code running behind a local proxy with a find-and-replace engine that swaps "Claude"→"Cursor" in system prompts and messages. They bundle the full @anthropic-ai/claude-agent-sdk and @anthropic-ai/claude-code packages, plus a custom fine-tuned model (claude-3.7-sonnet-finetuned-cursor-20250514-v1)" Full report gist.github.com/jasonkneen/4c0…
中文
We're delighted to announce that MiniMax M2.7 is now officially open source.
With SOTA performance in SWE-Pro (56.22%) and Terminal Bench 2 (57.0%).
You can find it on Hugging Face now. Enjoy!🤗
huggingface：huggingface.co/MiniMaxAI/Mini…
Blog: minimax.io/news/minimax-m…
MiniMax API: platform.minimax.io
English
Lovable CEO realizing their biggest model provider is now their biggest competitor.
can@marmaduke091
Another leak from Anthropic They created a lovable-like feature where you can build full-stack apps easily They are coming after everthing
English
New in agent-device: image diffs are now visual + machine-readable.
Agents can tell:
→ where the change happened
→ text movement vs visual change
→ which part of the screen changed
From “close enough” to “Wi-Fi moved 17px”.
This cuts design→code iterations by 40–60%.
English
Jason@JasonBoy7
English
@addyosmani How to use the cli? did not find anything related in the doc?
English
Want to give your agent quality checks?
Chrome's DevTools MCP now includes:
⚡️ Performance checks via Lighthouse
📈 Memory leak detection Skill
🦻 Accessibility debugging Skill
🎨 LCP optimization Skill
and an experimental new CLI 👀
English
@MinLiBuilds /compact 小于100K一般也不会用吧，如果TTL超了，相关度也不大了，手动compact一下是不是比等着自动compact更好点；官方好像还对compact做了caching优化
x.com/trq212/status/…
Thariq@trq212
中文
