Meta open-sources design system Astryx under the MIT license, with accompanying CLI and MCP
Meta has open-sourced Astryx, a design system refined internally over 8 years and used in more than 13,000 applications, under the MIT license, and it has already gained 27.4K stars. The system includes over 150 components such as buttons, forms, modals, dark mode and brand themes, and natively comes with a CLI and the MCP protocol, so AI can directly read its component documentation and standard templates when writing code. The author believes this enables tools such as Cursor and Claude Code to assemble frontend code according to unified specifications rather than improvising; Astryx is currently still in Beta.
Why read this
AI score 62
The original article points out that this design system natively comes with a CLI and MCP, so readers can judge whether AI can obtain reusable specification constraints when writing frontend code.
Topics
View original postFully open source!! I used Tencent WorkBuddy to run a complete AI investment research workflow for Hong Kong and US stocks: reviewing holdings, breaking down financial reports, and relentlessly tackling data hallucinations, with a 4-step prompt included—I suggest copying it directly. As long as you regularly read financial reports and manage holdings, you can directly use this human-AI collaboration as an AI investment research SOP workflow for investment research reviews going forward: the core logic is not to have AI tell fortunes and guess tomorrow's gains or losses, but to treat it as a cold auditor that thoroughly verifies the chain of evidence behind every number. Perhaps only after doing this calculation will everyone understand how important this system is: What you fear most when reading financial reports is never getting the big trend wrong, but AI fabricating a fake data point, or mixing in the company's restated accounting basis for the second year, and your real-money decision involving tens of thousands or even hundreds of thousands is instantly ruined; And this workflow illuminates all the hidden pitfalls, so even if you lose money, you can review it clearly! I have always felt that what AI should really do in investment research is the most labor-intensive, dirty and tiring work. I summarize it as conducting a background check on numbers, every data point must pass three checks: → Where it came from: which section of the annual report, and clicking it must trace back to the original text → What it is compared with: whether the cross-period accounting basis is the same, and whether they can be directly subtracted → Who said it: whether it is a verified fact or the management's own explanation. No number whose background cannot be clarified is allowed into the decision conclusions. The video contains the complete screen recording (all demonstrations use anonymized synthetic data), and the entire test is divided into four extremely clean steps: Build a holdings dashboard, and never guess blindly when cost is not provided. Drop in the raw holdings and price snapshot, and convert weights into Hong Kong dollars by currency. For positions without recorded cost, honestly mark the cost as missing, and eliminate model hallucination interpolation. Dynamic concentration risk control. After adding purchase records, automatically recalculate market value. When a single position is diluted to 29.18% due to adding to the position, the concentration warning is automatically withdrawn, and the before and after of the rebalancing are clear at a glance. Investigate the basis traps of annual report restatements. Break down business segment growth, and when the company restates its basis across periods, the numbers are precisely anchored to the original annual report section, and clicking opens the restatement notes, avoiding misleading fake increments. Ultimate follow-up, three-column physical isolation. When accounts receivable show unusual movements, force it to break them down into three columns: verified facts, the company's own explanations, and judgments still requiring manual verification, completely isolating management's pie in the sky from objective facts. Finally, I want to tell everyone that investment research tools are never crystal balls; what is more terrifying than losing money is losing according to no system of your own. Let's encourage each other! The original 4-step prompt and experience guide are organized in the comments section; as usual, help yourselves👇
View original
全盘开源!!我用腾讯 WorkBuddy 跑通了一套港美股 AI 投研工作流:复查持仓、拆解财报、死磕数据幻觉,附 4 步提示词,建议直接照着抄。 只要你平时看财报、管持仓, 以后做投研复盘可以直接拿这套人机协作当AI投研sop工作流:核心逻辑不是去让 AI 算命猜明天涨跌,而是把它当成冷酷的审计员,把每一笔数字背后的证据链彻底查清。 可能算过这笔账大家才知道这套体系有多重要: 平时看财报最怕的从来不是看错大趋势, 而是 AI 瞎编一个假数据, 或者把公司第二年重列的会计口径混着算, 你几万甚至几十万的实盘决策就当场踩雷; 而这套工作流把所有暗坑全部照亮, 亏了钱也能清清楚楚复盘! 我一直觉得AI 在投研里真正该干的就是最费人工的脏活累活。 我把它总结为给数字查户口, 每一笔数据都必须过三关: → 哪来的:出自年报哪一节,点开必须能追溯到原文 → 跟谁比:跨期会计口径一不一样,能不能直接相减 → 谁说的:是已经核实的事实,还是管理层自己的解释 户口查不清的数字,一个都不准进决策结论。 视频里是完整录屏(演示全部使用匿名合成数据), 整套实测分为极度干净的四步: 做持仓看板,没给成本坚决不瞎猜 丢入原始持仓和价格快照,按币种折算港币权重。遇到未录入成本的头寸,老老实实标出成本缺失,杜绝模型幻觉插值。 集中度动态风控 补录买入记录后自动重新折算市值。 单只持仓因加仓被稀释到 29.18% 时, 集中度预警自动撤销,调仓前后一目了然。 排查年报重列的口径陷阱 拆解业务分部增长, 遇到公司跨期重列口径时, 数字精准锚定在原始年报章节, 点开就能看到重列附注,避免假增量误导。 终极追问,三栏物理隔离 遇到应收账款异动, 强制要求它分三栏拆解: 已验证事实、公司自己的解释、还需人工核实的判断,把管理层画饼与客观事实彻底隔绝。 最后想跟大家说,投研工具从来不是水晶球,比亏钱更可怕的是没有按照自己的体系去亏,共勉! 4 步提示词原文与体验指引整理在评论区了,老规矩大家自取👇
Body · AI translation
Holy crap, Meta has really pulled out its long-hidden treasure this time. Astryx, a design system it refined internally for a full 8 years and got running across more than 13,000 apps, has been open-sourced directly under the MIT license, and it already has 27.4K stars🌟!!
Honestly, if it were just another ordinary open-source UI component library, even one stuffed with 150+ ready-made buttons, forms, modals, dark mode and brand themes, I wouldn't be this excited.
After all, the frontend world has never lacked wheels. What really made my eyes light up is that Meta took the appetite of AI Agents into account this time.
This system natively ships with a CLI and the MCP protocol, so when AI writes code it can directly read its component docs and standard templates.
Anyone who has done Vibe Coding surely knows the feeling. Back when you had Cursor or Claude Code write frontend code, the functional logic ran blazing fast,
but when it came to visual and interaction details, it was all down to luck and gacha pulls—ten revisions could give you eight feral styles, and in the end you still had to tweak the styles yourself bit by bit.
What's interesting is that once this kind of design spec, validated across tens of thousands of big-company apps, is fed directly into the AI's context via MCP,
the nature of writing frontend changes completely. AI is no longer blindly guessing and doodling; instead it's put inside an industrial-grade aesthetic guardrail and assembles directly according to standard blueprints.
I've always felt that what frontend development lacks most in the AI era is never the speed of generating code, but this kind of deterministic spec that can be fed directly to models.
Although Astryx is still in Beta, following this path, the visual floor for one-person companies hand-building products
is probably about to be pulled straight up to parity with first-tier teams by this wave of big-company infrastructure.