推文
760 tweets · 188 sources账号:全部mattpocockukdair_aisimonwbchernyadocompletetrq212dexhorthyyetonefelixrieseberg0xblacklightdani_avila7zarazhangruiAlchainHustbadlogicgamesdoteypetergyangvikingmuteryancarsonClaudeDevskunchenguidleon7haoctatedeviamzhihuikieranklaassenmckaywrigleylennysanthdxrRhysSullivangarrytanjakevin7johnlindquistkarpathyservasyy_aitheo0xMovez9hillsRLanceMartinaidenybaidingyihylarucodermitchellhmitsuhikomvanhornomarsar0steipetezeegGeminiAppGergelyOroszZaynHaoaddyosmaniantirezdanshipperdillon_mulroyelonmuskewind_devidoubiccnummanalirealWeZZardswyxthsottiauxyiliush0xPaulius0x_rodyAYi_AInotesAnthropicAIBarret_ChinaBenJames_____CMGS1988ChadMoranDanielMiesslerDaveJDavidKPianoDimillianFactoryAIFardeemMFarzaTVGoogleDeepMindHamelHusainHiTw93HilaShmuelIfanJewIndieDevHaileyJack_W_LindseyJackywineJasonZXJiaxi_CuiJinjingLiangKhalidWarsaLinearUncleLinghuaJMatthewBermanMaxForAIMeari_V2_0_GMnilaxQingQ77RaillyHugoSaccc_cTaniyatweets_VincentLogicai_explorer25aibuilderclub_alexalbert__alliekmillerandrewfarahantoinecojpanxue201arkuy99artmanasynkimobbssppllvvbearliubentossellbridgemindaibuaaxhmcailynyongyongcgtwtscharmaine_kleechrisbarberchrisparkXclairevodelba_oliveirademishassabisdoodlesteindriaforallelithrarelvissunericzakariassonfrancoisfleuretgabriell_labiamsahaj_xyzimdigitalashishimvihvjarredsumnerjerryjliu0jianshuojiayuan_jyjlongsterjonas_nellejoshalbrechtjpschroederkaiofreitaskepanolexruslogancyangmattypnabeelnifinetninthbit_aioran_gepatrickcpejmanjohnpetradonkaprathamgrvprathyvshquanruzhuoxiuquant_sheepraroquerealCaigurepsiacerohanpaul_aisamasamuelstroscheisawyerhoodshadcnshao__mengshaogefenhaoshreyansjtechgirl1908techwith_ramthorstenballtianyitobiaswuptoddsaunderstricaltturingoututuretomuniswap12untraceable_thevanillaCitronvelvet_sharkwengtianxinwquguruxiaogaifunxicilionycombinatorzachlloydtweetsziwenxu_zodchiii
@dani_avila7·6月19日For You
New Claude Code (v2.1.181) has a small but useful DX improvement
You can now set any session setting inline with /config key=value, without opening the menu:
/config useAutoModeDuringPlan=false
Works in interactive, -p, and Remote Control
Would like the same pattern for /model, /effort, /color, and even skill arguments 👀
新版 Claude Code (v2.1.181) 有个小而实用的 DX 改进： 现在可以用 /config key=value 内联设置任意 session 参数，不用再打开菜单： /config useAutoModeDuringPlan=false 在 interactive、-p 和 Remote Control 模式下都有效。 希望 /model、/effort、/color 甚至 skill 参数也能用同样的模式 👀
@dani_avila7·6月16日For You
Planning decisions vs execution decisions using Claude Code
You decide what to do, claude decides how: 70% planning by the user, 80% execution by claude
In my experience the more you know about your domain, the better your results
A lawyer can write excellent plans and Claude executes them. Same with a software developer, the better you are, the better software you ship 🚀
用 Claude Code 区分「规划决策」和「执行决策」 你决定做什么，Claude 决定怎么做：70% 的规划由用户完成，80% 的执行交给 Claude。 我的经验是：你对自己领域越熟悉，结果越好。 律师可以写出优秀的方案，Claude 来执行。软件开发者同理——你越强，你做出的软件就越好 🚀
@dani_avila7·6月4日For You
Progressive disclosure in Skills doesn't get enough attention, and it's probably one of the most powerful patterns for documenting large agent processes
Script execution and file references are extremely powerful, especially now that LLM costs keep climbing
Token optimization is becoming a science. If you want to get serious with agentic systems, you need to prioritize it
3 stages to understand how it works and start using it in your Skills:
Stage 1 - Startup: Claude scans all skills, loads only the YAML frontmatter. ~60 tokens per skill. The agent knows what it CAN do, not HOW
Stage 2 - Task match: User request triggers intent recognition. Only the matching SKILL.md body loads. ~1,500-2,000 tokens, everything else stays on disk, zero context cost
Stage 3 - Deep execution: Reference files and scripts load on demand. Scripts run without their source ever hitting the context window
The math: 8 skills, 70K tokens of knowledge, ~500 tokens at startup
That ratio is why this pattern matters
Skills 里的渐进式披露（progressive disclosure）没有得到足够重视，它可能是文档化大型 agent 流程最强大的模式之一。 脚本执行和文件引用极其强大，在 LLM 成本持续攀升的今天尤为如此。 Token 优化正在成为一门科学。如果你想认真做 agentic 系统，必须把它放在优先级上。 理解它的工作原理，分三个阶段： 阶段 1 - 启动：Claude 扫描所有 skills，只加载 YAML frontmatter。每个 skill 约 60 tokens。agent 知道它「能做什么」，而不是「怎么做」。 阶段 2 - 任务匹配：用户请求触发意图识别，只有匹配的 SKILL.md 正文被加载。约 1500-2000 tokens，其余内容留在磁盘，零 context 开销。 阶段 3 - 深度执行：引用文件和脚本按需加载。脚本运行时，其源代码甚至不会进入 context window。 算一下：8 个 skills，70K tokens 的知识库，启动时只用 ~500 tokens。 这个比例，就是这个模式重要的原因。
@dani_avila7·5月18日For You
At my previous startup we tried RAG with semantic search and vector DBs on large codebases
The index was always one sprint behind the repo
What actually works... invest in the harness and let the agent traverse the live codebase https://t.co/8zlIbtQdt6
@dani_avila7
Read it before the weekend's over! Every enterprise Claude Code rollout hits the same wall: the model gets better, the setup doesn't, nobody owns it Anthropic just named the role that fixes that: Agent Manager... Owns the CLAUDE.md hierarchy, the plugin marketplace, permissions, and which skills ship
我之前的创业公司在大型代码库上试过 RAG + 语义搜索 + vector DB。 索引永远落后代码库一个 sprint。 真正有效的做法：投入到 harness 本身，让 agent 直接遍历实时代码库。 --- 引用推文：趁周末结束前读一下！ Claude Code 的每次企业级推广都撞上同一堵墙：模型越来越强，搭建方式没跟上，没人负责。 Anthropic 刚刚给解决这个问题的角色命名了：Agent Manager…… 负责 CLAUDE.md 层级体系、plugin 市场、权限管理，以及哪些 skill 上线。
@dani_avila7·5月13日For You
I'm not using "claude agents" but Claude is still splitting the plan's tasks across different agents
I can switch between each agent's session just by selecting it
All running in parallel... how is this different from claude agents? 🤔 https://t.co/jlJ492jD1w
我没有使用「claude agents」，但 Claude 还是把计划里的任务拆分给了不同的 agent。 我可以通过选择切换每个 agent 的 session。 全部并行跑着……这和 claude agents 到底有什么区别？🤔
@dani_avila7·5月11日For You
Claude Code 2.1.139 added /goal
You set a completion condition and Claude keeps working across turns until it's met
Works in interactive, -p, and Remote Control 👏 https://t.co/ETyGjplaOC
Claude Code 2.1.139 新增 /goal 命令。 你设定一个完成条件，Claude 会跨多轮持续工作直到满足条件。 支持交互模式、-p 参数模式和 Remote Control 模式。
@dani_avila7·5月9日For You
ALWAYS review and clean up MCPs in every Claude Code session
Not a common practice, but it should be
And if you're building Skills, declare the tools in the frontmatter using the allowed-tools field
In a session with many MCPs connected, Claude can pull tools from any of them mid-execution, even ones unrelated to what the Skill is doing
Scoping the tools keeps the Skill predictable and avoids tool calls leaking in from MCPs that happen to be loaded
每次 Claude Code session 都要记得检查和清理 MCP，这不是常见做法，但应该成为习惯。 如果你在构建 Skills，记得在 frontmatter 里用 `allowed-tools` 字段声明工具。 在连接了多个 MCP 的 session 里，Claude 在执行过程中可以从任意 MCP 拉取工具——包括那些跟当前 Skill 完全无关的。限定工具范围能让 Skill 行为可预测，避免不相关 MCP 的工具调用悄悄渗透进来。
@dani_avila7·5月8日For You
This functionality inside Claude Code Desktop is great
Being able to see the files while it works has me ready to drop VSCode in a few weeks... but!
It would be way better if it showed git colors inside the files. That way you'd know which ones have been edited within the commit
VSCode is now basically a pre-commit diff viewer. If this can be done and there's some way to enable these green and red files to show the git diff, please tell me how!
CC @amorriscode
Claude Code Desktop 里的这个功能很棒。 能在它工作时实时看到文件，让我几乎准备好几周后抛弃 VSCode……但是！ 如果文件里能显示 git 颜色就更好了。这样你就能知道哪些文件在这次 commit 中被编辑过。 VSCode 现在基本成了我的 pre-commit diff 查看器。如果 Claude Code 能做到这点、能让这些绿色红色的文件显示 git diff，请告诉我怎么开启！
@dani_avila7·4月7日For You
Just finished watching the "What We Shipped" webinar from the Claude Code team with @trq212 and @dmwlff
Worth saving and watching in full when you have time
Good overview of everything the team has been shipping lately
Webinar: https://t.co/5VMHJSkzev https://t.co/3aq37bnJHw
刚看完 Claude Code 团队的「What We Shipped」webinar（主讲 @trq212 和 @dmwlff） 有时间的话值得完整看一遍 对团队最近发布的所有内容做了很好的梳理 Webinar 链接：https://t.co/5VMHJSkzev
@dani_avila7·4月6日For You
There's a lot of confusion between Skills and Subagents in Claude Code
And it gets worse when you realize a Subagent can use Skills, and a Skill can spin up Subagents
Wrote an article breaking down how each actually works and, more importantly, when to reach for one vs the other
@dani_avila7
Skills can use subagents, Subagents can use skills
Claude Code 里的 Skills 和 Subagents 让很多人困惑 当你意识到 Subagent 可以调用 Skills、Skill 也可以启动 Subagents 时，混乱更甚 写了一篇文章拆解两者实际如何工作，更重要的是——什么情况下该用哪个 [引用] Skills 可以使用 Subagents，Subagents 可以使用 Skills
@dani_avila7·3月30日For You
Cohere released a new Transcribe model, so I built a Chrome extension to test it
It works two ways: through the API or with a local server
API mode: grab a free key from your Cohere account, https://t.co/gnf0i2Ul2s sign up and you get free tier access
Local mode: download the model from Hugging Face, https://t.co/V6RJZzmWC9, spin up the local server and add your HF access token from here: https://t.co/ZYIufCcxbg
Pick either mode and you're good to go, left it open source under MIT: https://t.co/ifZLkrMFuK
Good weekend project, going to keep exploring where else this can be applied
Thanks @nickfrosst and the @cohere team!
@nickfrosst
@cohere transcribe Sota open source transcription model running in the browser :) Weights on @huggingface link below https://t.co/OmrHFA94lG
Cohere 发布了新的 Transcribe 模型，于是我做了一个 Chrome 扩展来测试它。 支持两种模式：通过 API 或本地服务器。 API 模式：从 Cohere 账户获取免费 key，注册后即可使用免费配额。 本地模式：从 Hugging Face 下载模型，启动本地服务器，并添加你的 HF access token。 两种模式任选其一，已以 MIT 协议开源。 不错的周末项目，接下来会继续探索它的其他应用场景。 感谢 @nickfrosst 和 @cohere 团队！ > 引用 @nickfrosst：@cohere transcribe —— 目前最优的开源转录模型，可直接在浏览器中运行 :) 权重已上传至 @huggingface
@dani_avila7·3月28日For You
In this Anthropic blog post about Claude Code Plugins https://t.co/UHKX0DJOj2
You'll find a link to my project https://t.co/4KGYSLvEP2
Inside there's a collection of plugin marketplaces I've been testing and reviewing
The site ranks them by GitHub stars and extracts the info to display it in a way that's easier to understand and use
I'm running a Scheduled Task that keeps searching for more CC marketplaces
If you have one, send it my way and let's keep building this Claude Code Templates community
在 Anthropic 这篇关于 Claude Code Plugins 的博客里，你能找到我项目的链接。 里面收录了我一直在测试和评测的插件市场合集。这个站点按 GitHub stars 排名，并提取信息以更易读、易用的方式展示。 我设置了一个定时任务，持续搜索更多 CC 插件市场。 如果你有自己的，发给我，一起把这个 Claude Code Templates 社区做起来。
@dani_avila7·3月25日For You
Claude Code Desktop is bringing GitHub Issues directly into its environment
The more this platform evolves, the closer it gets to GitHub...
Connect your PRs straight to CI for every deploy and you've got Claude shipping different versions of your app ready for QA https://t.co/RSMePsdqJG
Claude Code Desktop 正在把 GitHub Issues 直接引入其环境。 这个平台越来越像 GitHub 了…… 把 PR 直接接入 CI 用于每次部署，你就有了一个 Claude 自动生成不同版本 app 供 QA 测试的工作流。
760 tweets · 188 sources