Peter

[CodeAlpha @RealCodedAlpha]

Anthropic 官方团队把 Claude Code 从底层原理到真实项目实战，完整讲了一遍！

如果你只是会让 Claude Code 写代码，却还没搞懂它如何读取文件、调用工具、管理权限、调度 subagents，这个视频值得收藏！

我整理了重点时间线：

00:00｜Claude Code 底层如何工作
04:51｜Permissions：哪些操作允许、询问或拒绝
10:46｜如何把 Code Review、部署流程做成 Skills
15:08｜Intent-driven development 到底怎么做
22:40｜用语音 Prompt 从零构建 3D 游戏
32:55｜用 Branch、Draft PR 和 GitHub CLI 控制交付
38:15｜Auto Mode 如何减少重复授权
40:48｜用 Dynamic Workflows 并行调度多个 Subagents
46:53｜从 Demo 走向 Production 的工程原则
55:33｜Claude Code 如何用于研究、数据和媒体处理

我认为最值得看的是 Lydia Hallie 拆解了 Claude Code 的真实运行方式：模型本身没有长期状态。

文件内容、终端输出、Git 分支、工具调用结果和会话历史，都是由外层的 Agent Harness 维护，再在每一轮重新组装给模型。

这也解释了为什么真正影响 Claude Code 效果的，不只是 Prompt，还有：CLAUDE.md、Skills、Permissions、Verification、Git Workflow、Subagent 编排！

视频最后还有一个很重要的提醒：AI 能一次生成大量代码，不代表这些代码都应该被合并。

架构、验证、安全、Review 和最终质量，依然由人负责。

这可能是目前理解 Claude Code 最完整的一小时。

建议先收藏，再跟着时间线看！

[OKX @okx]

Welcome to the next generation of digital finance in Europe 🇪🇺

OKX is MiCA-authorised, offering a single unified account with institutional-grade infrastructure and deep liquidity.

Switch now: 8% deposit bonus + €400 new user reward. €10M in rewards up for grabs.

[Peter @lunacyyan]

https://t.co/mGNg2vr6VG

[Star_OKX @star_okx]

The end of regulatory arbitrage is good for crypto.

Competition should be based on products, technology, execution, governance, and trust — not on who can operate with the fewest rules.

That is better for users, better for the industry, and ultimately better for innovation.

[Star_OKX @star_okx]

https://t.co/64sA9O0D0T

[Ren @Ryrenz]

https://t.co/vaAFLJB2jp

[Star_OKX @star_okx]

Tokenized equities are one of the most important RWA use cases.

We expect to see xStocks on X Layer soon, bringing global equities onchain and expanding access to capital markets.

Access to global markets should be as open as access to information.

[Soran @Soranlan]

https://t.co/iOxOaw2wD1
Anthropic 工程师一句话，直接撕开了大多数人的工作幻觉：

你不是来盯着 Claude Code 一行行干活的。
你该做的只有一件事

睡醒，验收它昨夜交付的成果。
然后更狠的来了。

她开着直播，22 分钟，从零搭完一整套自动工作流。

扎心对比现在才出现：
大多数人一关终端，任务就死了。
而这套配置，在你睡觉的时候继续跑、继续改、继续交付。
视频放下面了。

看完你可能会忍不住问自己
我干嘛还守着屏幕当人肉监工？

[Star_OKX @star_okx]

In the agent economy, APIs won’t just authenticate users — they will negotiate, charge, and settle autonomously.

[Star_OKX @star_okx]

Happy to see CoinAnk integrating with our Agentic Payment Protocol.
APP is designed to become the future payment channel for API service providers in the agent economy.

[Star_OKX @star_okx]

Looking forward to seeing xStocks on X Layer, bringing tokenized equities and RWA assets into the ecosystem.

[xStocks @xStocksFi]

Onchain markets are opening up.

@XLayerOfficial Exchange OS lets any builder deploy spot, perpetuals, or outcomes markets.

Open-protocol, permissionless infrastructure.

Tokenized equities and ETFs are part of what comes next.

[Centrifuge @centrifuge]

Market infrastructure as a shared layer instead of something every venue rebuilds.

The same vision Centrifuge has been pushing for since 2017.

Joining Exchange OS to push it further: distribution, capital efficiency, and the next wave of composability for tokenized assets.

[Star_OKX @star_okx]

Build together. Scale together.

[Akshay 🚀 @akshay_pachaar]

from prompt to context to harness engineering.

three terms keep coming up in AI engineering, and they get conflated all the time. here is the cleanest way to understand what each one is and how they fit together.

𝗽𝗿𝗼𝗺𝗽𝘁 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗶𝘀 𝘁𝗵𝗲 𝗺𝗲𝘀𝘀𝗮𝗴𝗲.

the model has no memory of anything before this single call, so the prompt has to carry the full universe of what it needs to know. that means a role, some background, the instructions, a few examples, and a format.

these get assembled into one input and sent to the model. when the output falls short, the skill is figuring out which ingredient is actually letting you down, not rewriting the instructions every time.

the unit of work is one input.

𝗰𝗼𝗻𝘁𝗲𝘅𝘁 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗶𝘀 𝘁𝗵𝗲 𝗺𝗲𝗺𝗼𝗿𝘆.

across multiple steps, the window is finite and the information available is not, which forces a curation step. without it, important details get buried under stale tool outputs and old turns, and the model's attention degrades on the things that actually matter.

a curator selects what stays, compresses what is useful but bulky, and drops the rest. each step's output then feeds into the next step, where good curation is more about knowing what to throw away than packing more in.

the unit of work is what stays in the window, step by step.

𝗵𝗮𝗿𝗻𝗲𝘀𝘀 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗶𝘀 𝘁𝗵𝗲 𝗺𝗮𝗰𝗵𝗶𝗻𝗲.

on its own, a model just generates text. the harness is what turns it into something that can take actions, check its own work, and recover when a step goes wrong.

the full loop has three phases:

- 𝗴𝗮𝘁𝗵𝗲𝗿 pulls together everything the model needs
- 𝗮𝗰𝘁 runs the model and calls tools or sub-agents
- and 𝘃𝗲𝗿𝗶𝗳𝘆 checks the output with tests or a judge

on failure, the whole loop retries with updated context, which is the entire difference between calling an API and running an agent.

the unit of work is the machine itself.

here is the part that ties it together.

prompt engineering and context engineering both live inside 𝗴𝗮𝘁𝗵𝗲𝗿. the harness is the outer container, context is what it curates, and the prompt is what it finally hands to the model.

zoom out and the unit of work gets bigger. zoom in and you are back at the prompt.

i also published this deep dive (article) on agent harness engineering, covering the orchestration loop, tools, memory, context management, and everything else that transforms a stateless LLM into a capable agent.

the article is quoted below.

[Charlie Hills @charliejhills]

One folder runs my entire content business.

Here is everything inside ~/.claude:

✦ CLAUDE .md → your employee handbook for Claude.

Every time you open a new chat, Claude reads this.
Before it says a word, it knows who you are.

Put in here:

- Your writing rules and tone
- Your banned words and phrases
- The projects you are working on
- The things Claude should never do

Without it, Claude is a stranger every session.
With it, Claude feels like it knows your work.

✦ CLAUDE.local.md → your personal overrides.

Your personal overrides on top of CLAUDE.md.
Never hits the team repo. Never shared.

The stuff that is yours alone:

- Preferences your teammates do not need to see
- Anything you want Claude to know about you
- Personal shortcuts and style rules

✦ settings.json → the control room.

Decides which tools Claude can use without asking.
Also holds your hooks registry.

☑︎ Default model selection
☑︎ Permissions allowlist
☑︎ Hooks registry

✦ commands/ → your slash shortcuts.

For example: /today, /create, /ideas.

200+ word prompts compressed into 8 characters.
You type one word. The whole workflow fires.

☒ No copy-pasting prompts
☒ No re-explaining yourself every session

✦ skills/ → keyword that trigger workflows.

Each one is a complete system Claude runs on demand.

Type "create" and it writes a post in my voice.
Type "infographic-maker" and it builds the graphic.
Type "ig-scripts" and it turns scripts into Reels.

This alone saves me 4 hours every single week.

✦ agents/ → where Claude stops being a chatbot.

I have 12 sub-agents in 2 agent teams.

- One does strategy
- One does design
- One builds
- One reviews
- One publishes

You brief the first. It hands off automatically.
You come back when the pipeline is done.

✦ hooks/ → scripts that fire on every tool use.

Before Claude reads a file, a hook checks it.
After it writes something, a hook reviews it.

These are your guardrails.

Claude cannot go off-script even if it tries.

✦ plugins/ → pre-built packs. One install.

Mine include:

- Vercel
- Telegram
- claude-mem
- frontend-design

These are bundled commands and MCPs.

✦ channels/ → bridges to iMessage and Telegram.

This is what makes it genuinely mobile.

Send a message from your phone.

- Claude fires the skill.
- Runs the entire workflow.
- Sends the output back to you.

No laptop, terminal or desk required.

✦ mcp.json → every connected tool in one file.

Mine has:

- playwright (so Claude can browse the web)
- figma (so it can read my designs)
- gmail (so it can check my inbox)

Every tool you connect here is one Claude can reach.

That is it. One folder.

Everything Claude needs to run your work.

Start with CLAUDE .md and nothing else.
Build the rest as you actually need it.

Full beginner guide here

https://t.co/GXvwuzHk5l

Repost ♻️ to help someone in your network.

[Movez @0xMovez]

Anthropic AI engineer just showed how to give AI agents real memory in 4 steps - and it changes everything

in 28 minutes he shows exactly how agents can remember across sessions, completely free

worth more than any $500 AI engineering course

here's what he covers:

• why agents forget everything between sessions
• memory stores - agents read, write across sessions
• dreaming - agents that improve their own memory
• 95% cache hit rate, so it stays cheap

most people are still copy-pasting context into every new chat - while the people who figured this out are building agents that get smarter every single night

watch full video then read article below

[Peter @lunacyyan]

https://t.co/7hXC4FlEin

[Tw93 @HiTw93]

https://t.co/x9eqpjrSuS

[Star_OKX @star_okx]

The upgrade of the financial system is already happening — built for internet-native users and AI agents.
OKX is building the New Money App for both.