Aslıhan Karaca

ONE DİRECTİON_LOUİS I LOVE YOU

[arc. @arceyul]

Este tipo le dio a Claude Code la habilidad de copiar el diseño de cualquier web.

Creó un MCP que lee el HTML/CSS de cualquier sitio y extrae colores, tipografía y componentes.

Resultado: UIs limpias por fin.

GUÁRDATELO 🔖

[Aslıhan Karaca @BerilTu1]

@zackpaid FABLE

[Rachel🥥 @Zesee]

https://t.co/bingMhkVQf

[Dave @DaveDiederen]

How to build high-quality advertorials & listicles fast with Claude

[Aslıhan Karaca @BerilTu1]

@maxxmalist AGENT

[Aslıhan Karaca @BerilTu1]

@jakobcounts123 SOP

[解药三分毒 @jieyaoSFD]

#BTC 目前熊市初期的下跌结构同2022年熊市初期相似。

熊市中第一段反弹也相似，都是在地缘冲突带来的不确定性中迷茫的挣扎。

一样的出身，一样的经历，你觉得会有相似的未来吗？

[zack @zackpaid]

i replaced 3 FULL-TIME roles with 2 AI tools + an MCP

> a researcher
> a creative strategist
> a media buyer pulling 8-hour shifts

COMPLETELY GONE….

Manus handles the intel layer. Claude Opus 4.8 + the Meta MCP handles execution. and the two of them running together is what lets me deploy $1,000,000+ a month in meta ad spend without a full team bleeding budget while i sleep.

i spent days documenting the ENTIRE system...

here's what's inside:

> the negative review mining workflow
how i send Manus into the 1-star and 2-star reviews of 4-6 competitors and pull out the buying angle gaps the market has already proven exist but nobody is running ads on (the fractional CFO campaign that came back with "the advisor who actually talks to you" off a single 20-minute research run, not a brainstorm session)

> the 20-25 cell creative matrix
how the Manus research output gets mapped to buying trigger, awareness stage, format type, angle category, primary objection, and opening line before a single word of copy gets written, so every ad in the batch is a genuinely different creative concept and not a variation the andromeda container system collapses into one slot

> the full Meta MCP setup
exact JSON config block, file path for both mac and windows, system user token setup with the right permissions (ads_management, ads_read, business_management), the node.js version requirement, and the verification command that confirms the connection is live (plus the two no-config paths if you don't want to touch a json file at all)

> the 4 execution jobs the Meta MCP runs automatically
creative performance mapping by hook category and format type in the first 72-96 hours, 15-minute budget reallocation with the specific ROAS thresholds that trigger a move, creative fatigue detection across 4 signals firing at once, and the campaign health log you wake up to instead of a dashboard

> the entire prompt library
copy-paste prompts for the daily 5-minute briefing, the monday deep-dive (90 minutes of spreadsheet work down to 8), campaign creation, signal diagnostics, and the monthly report that pastes straight into a client deck

> the multi-client agency setup + the time math
how to wire one claude project per client so the same prompts work across the whole roster, and the actual numbers on an 8-client book (16-24 hrs/week of account management down to 2.5-3 hrs, 14-21 hours back every week)

and MUCH more inside…..

all backed by $10,000,000+ generated for service-based businesses through meta ads funnels.

like + comment "META" and i'll send it over

(must be following + RT for priority access)

[Aslıhan Karaca @BerilTu1]

@zackpaid META

[Aslıhan Karaca @BerilTu1]

@eCom_Amin MANUS

[Amin @eCom_Amin]

manus + opus 4.8 are CRACKED at oneshotting google ads funnels for ecom…

> advertorials
> comparison pages
> quizzes, listicles, even pdps

so i put together a 6-step playbook on how we're ABUSING it to maximize every click from our clients’ google ads campaigns

covering EVERY prompt, template, and principle:

> keyword clustering for TOF/MOF/BOF
> funnel type selection guidelines (adv vs VSL vs pdp)
> AI copy generation based on traffic’s intent
> google ads campaign architecture
> audience signal library
> pre-launch audit

all backed by having generated +$20M for ecom brands with google ads funnels like these

like + comment "MANUS" and i'll DM you EVERYTHING

(must be following + RT for priority access)

[SilenceÇaPrompt @SilenceCaPrompt]

TU PEUX TÉLÉCHARGER N'IMPORTE QUEL SITE WEB EN ENTIER, ASSETS COMPRIS.

Website Downloader est désormais open source.

Il récupère le code source complet de n'importe quelle page, puis convertit les liens pour que tout marche sans connexion.

5 choses qu'il gère pour toi:

→ HTML, CSS et JavaScript
→ images et polices
→ conversion des liens en local
→ compression automatique en ZIP
→ envoi du fichier en temps réel

Le workflow qui te prenait un après-midi tient maintenant dans une seule interface.

Idéal pour cloner un projet, archiver un site ou apprendre du code des autres.

Regarde le repo en com👇

[Carl Weische @CarlWeische]

https://t.co/Xzm63hZerL

[Pepesso @0xPepesso]

🚨 $BTC WYCKOFF IS PLAYING OUT EXACTLY AS I CALLED IT

Three phases already completed:

1. SC (Selling Climax) - panic candle, volume spike, sharp capitulation drop

2. AR (Automatic Rally) - fast bounce straight off the climax low

3. ST (Secondary Test) - lower-volume retest of the lows, supply exhausted

One phase left: the Spring near $49k

The Spring isn't a crash - it's a classic market trap

Price dips below the accumulation range, shakes out every remaining weak hand, then reverses before anyone can react

The crowd sees a breakdown, Wyckoff traders see the entry

This is the mechanism: most people sell at the Spring, which is exactly why it has to happen - weak hands out before smart money can run price

Every independent signal is pointing to the $49K level

This is where patience pays

Follow + notifs on, I will keep you updated

[meng shao @shao__meng]

/writing-great-skills
https://t.co/UQbWpcuK3E

来自 152K✨ Skills For Real Engineers 作者 @mattpocockuk 的新 Skill，教咱们用最少但最有行为牵引力的结构，把 Skill 写成能稳定触发、分层加载、清楚完成、持续删减的“可预测工作流”。

# 跟这个优质 Skill 学它的编写思想

1. Skill 的根本目标是过程可预测
Skill 不是知识库，也不是提示词堆叠。它的作用是让模型在某类任务中形成稳定行为路径。好的 Skill 应该减少“这次做得细、下次做得浅”的波动。

2. 触发方式有成本权衡
它区分两类 Skill：
· Model-invoked：模型能自动发现并调用。优点是无需用户记住，缺点是 description 会长期占用上下文注意力。
· User-invoked：只有用户点名才会触发。优点是零上下文负担，缺点是用户必须记得它存在。

这里很关键：不是所有 Skill 都该自动触发。只有当模型确实需要自己识别任务，或其他 Skill 需要调用它时，才值得让它 model-invoked。

3. description 是触发器，不是简介
对于 model-invoked skill，description 的职责不是介绍得完整，而是准确告诉模型“什么时候该用我”。因此它应当前置关键触发词，只保留真正不同的触发分支，避免同义重复。

这点很实用：很多 Skill 写坏，是因为 description 像产品简介，而不是调用条件。

4. 信息层级决定 Skill 是否清爽
它提出一个三层结构：
· SKILL.md 中的步骤：模型必须按顺序做的事。
· SKILL.md 中的参考：模型运行时需要随手查看的规则、定义、事实。
· 外部参考文件：只在特定场景需要加载的材料。

好的 Skill 不把所有东西塞进主文件，而是用 progressive disclosure：常用、必须、影响流程的内容留在主文件；分支性、解释性、定义性内容放到外部文件，通过明确指针调用。

5. 每个步骤都要有完成标准
它特别强调 completion criterion。一个步骤不能只写“分析清楚”“完成检查”这种模糊目标，而要让模型能判断“是否已经完成”。

完成标准越清楚，越能防止模型提前进入下一步，也就是它说的 premature completion。

6. 拆分 Skill 不是为了整洁，而是为了控制注意力
什么时候拆？
· 如果一个 Skill 有独立触发词，可拆成单独的 model-invoked skill。
· 如果后续步骤会让模型急着往前跑，可把流程拆开，隐藏后续步骤，迫使模型认真完成当前阶段。

这很像工作流设计中的“减少提前优化”和“控制认知视野”。

7. leading word 是压缩行为的关键词
它提出一个很有洞察的概念：leading word。也就是用模型预训练中已经熟悉的强概念，来压缩一组行为要求。

例如与其反复写“快速、确定、低开销”，不如找到一个更有行为牵引力的词。好处有两个：节省 token，并且更容易稳定唤起模型已有的行为模式。

但它也提醒：弱词可能无效。例如“be thorough”如果只是模型默认会做的程度，那就是 no-op；需要更有约束力的词。

它的失败模式诊断很有用
这个 Skill 给出的几个常见问题非常精确：
· Premature completion：模型过早认为当前步骤完成。优先修正完成标准，而不是马上拆 Skill。
· Duplication：同一个意思出现在多个地方，增加维护成本，也会让某个概念被模型过度重视。
· Sediment：旧内容沉积，没人敢删，导致 Skill 越来越脏。
· Sprawl：内容都有效，但主文件太长，注意力被稀释。
· No-op：看似有用，实际不会改变模型行为的句子。

其中最有操作价值的是 no-op 测试：一句话如果删掉后模型行为几乎不变，它就不该留在 Skill 里。

[Amto @XAMTO_AI]

姚老师开源的 yao-meta-skill 我用下来，客观说，比官方那个 skill-creator 顺手太多了。

你只需要把那些散落的流程、提示词、甚至聊天记录一股脑丢进去，它就能帮你生成一个质量相当不错的 skill。

这项目是怎么来的？据说是扒了 Anthropic 泄露的 Claude Code 源码，又把各家模型的 skill 最佳实践揉在了一起，姚老师前后打磨了整整一个月。坦白讲，这是我目前用过最趁手的 Meta Skill，没有能打的。

🔗 https://t.co/YpDyTTWrDe

[philarekt @philarekt]

🚨 ONE MOVE LEFT BEFORE THE MARKUP BEGINS

Bitcoin is running the Wyckoff playbook exactly as written

SC, AR, secondary test all three already printed

Only one phase remains the Spring near $49K

The shakeout designed to break the last weak hands before the real move to six figures begins

The crowd is supposed to panic here that's the entire mechanism

Most people will sell exactly where they should be buying

I've called every major top and bottom for 15 years, including the $16K bottom and the $126K top both publicly, both before they happened

The next call will be even more important I'll post it here publicly like I always do

Turn notifications on, if you're not following yet, you'll understand why that was a mistake later

[Carl Weische @CarlWeische]

https://t.co/PBNAU82Kgn

[Lorenzo | Meta Ads & Performance Creatives 📈 @lorenzo_pravata]

I've just given away 10 prompts to create native ads from scratch. Here's how they work:

[Amin @eCom_Amin]

https://t.co/7Wnx7OKhQB