K T Tzy.

LLM

[alphaXiv @askalphaxiv]

Agents suck at making figures. They make things looks ugly and emphasize the wrong things.

We’ve made skills so your agents can create beautiful figures for your papers or visualizations while you are doing research.

GPT-5.6 Sol (Codex) w/o and with skills:

[lauren @poteto]

I'm writing a guide to pstack! Here's part one. https://t.co/4yjdWKPieE

[Machina @EXM7777]

here's every AI tool i use... and how they all fit together

the models:

- daily driver: GPT-5.6 Sol Low
- planning: Fable High
- coding: GPT-5.6 Sol High
- research: Opus 5 High
- image and video vision: Gemini 3.7 Flash
- design: Kimi K3 and Fable Low
- voices: ElevenLabs v3
- images: GPT Images 2
- video: Seedance 2.5 and Minimax H3

Claude and ChatGPT models run inside their own harnesses on max and pro subs, everything else goes through the API

- research: Parallel, Perplexity, Firecrawl, /last30days and Apify
- seo: Ahrefs, DataForSEO, Screaming Frog, Google Search Console and Porkbun
- emails: SmartLead and Zapmail
- analytics: Google Analytics and Posthog
- the rest of the stack: Browser Use, Typefully, Kit, Supabase and Vercel

how it runs together:

Hermes or Orca as orchestrators... they spawn new worktrees and Claude Code or Codex sessions inside them

planning loop: Fable writes the initial plan with /grill-me from Matt, then GPT-5.6 Sol workers roast that plan until it holds up

knowledge lives in Obsidian, memory across every agent and platform goes through mem0

all of it runs on a Hetzner AX42-1 box with proper backups and security

[K T Tzy. @KTTzyyy]

@gippp69 Just curious is there any tutorial to make something like that in the video?

[K T Tzy. @KTTzyyy]

@louszbd I use GLM-5, GLM-5.1, GLM-5.2 and currently using GLM-5.3 Flash. Overall it's great

But what I like more is the ZCode harness, freaking gooddd

I've been using ZCode for more than 2 weeks and NEVER back to OpenCode/OpenChamber/Cursor/Codex (Codex is better but much heavier), CC

[Vaishnavi @_vmlops]

A FREE GUIDE THAT COVERS EVERY PROMPT ENGINEERING TECHNIQUE YOU NEED

dair-ai's Prompt Engineering Guide is one of the most complete open resources on the topic 77k+ stars and still growing.

→ zero-shot, few-shot, and chain-of-thought prompting
→ ReAct, tree of thoughts, and self-consistency
→ RAG, function calling, and synthetic data generation
→ risks: adversarial prompting, bias, factuality
→ model-specific guides for GPT-4, LLaMA, Gemini, Mistral, and more

available in 13 languages, with notebooks and a full lecture included

https://t.co/aPhFQNTuxA

[Vox @Voxyz_ai]

Codex Sol has one bad habit: it turns small tasks into big systems.

Ask it to make a config change, and it builds a framework. Ask it to fix a bug, and it adds a new adapter layer.

Full text in the first reply. Add these rules to your AGENTS.md. 👇

[Suraj Sharma @suraj_sharma14]

For people who keep asking what to build in AI Engineering.

➣ Build your own Agent Orchestrator (deterministic state machine, no LangChain)

➣ Build your own Context Assembler (token-budgeted memory + retrieval + tools)

➣ Build your own MCP Server and Client (raw JSON-RPC, no SDK)

➣ Build your own Retrieval Stack (chunker + BM25 + dense + reranker from scratch)

➣ Build your own Eval Harness (trajectory grading + CI regression gates)

➣ Build your own Model Router (cost/latency/quality routing with fallbacks)

➣ Build your own Semantic Cache (embedding similarity + hit-rate tracking)

➣ Build your own Guardrails Middleware (injection detection + PII redaction)

➣ Build your own Streaming Proxy (SSE with TTFT and ITL metrics)

➣ Build your own Durable Workflow Engine (checkpoint/resume, mini-Temporal)

➣ Build your own LLM Tracer (OpenTelemetry-style spans for every hop)

➣ Build your own Sandboxed Tool Executor (isolated execution, resource limits)

➣ Build your own Prompt Registry (versioning + A/B routing + rollback)

➣ Build your own Data Flywheel (feedback → synthetic data → LoRA loop)

➣ Build your own Multi-Agent Consensus (weighted voting + judge + escalation)

Pick 3. Build them from scratch. Document every decision.

Most people import libraries.
Builders understand them.

(Bookmark & Repost.)

[Manish Kumar @Manixh02]

If you are building a vibe-coded app, you need these 3 websites;

1. Free-for dev;
https://t.co/8tXNuopjzu - A huge collection of free tiers, free developer tools, APIs, hosting, databases, and other resources.

2. Free Public APIs;
https://t.co/6wmzA8SERD - Find free public APIs you can use to add real functionality to your app without building everything from scratch.

3. OpenDesign;
https://t.co/3sNIuLNiFt - A collection of free, open-source design resources that can help you build better-looking interfaces faster.

Save this for your next vibe coding project 🔖

[Vuk.Digital @v4rvl]

Since people asked, a proper intro to the skill.
https://t.co/AdGbip9IUB

GPT models, Sol especially, love to overengineer process: hashes, locks, receipts, status files. Sooner or later the agent starts redoing finished work because a marker went stale, or refuses to act because its paperwork is missing, and reports all of that as progress. So you end up burning tokens and wasting time while accomplishing absolutely nothing.

The skill makes the agent classify every action as implementation, validation, or bookkeeping, and skip the third. Redoing work needs a real reason, and stale metadata is not one. Tests and benchmarks stay mandatory.

Result: the agent ships instead of managing its own paperwork. For me that took the experience of using GPT models from 3 to 8. I get stuff shipped and save on tokens.

The skill is generic and cuts this garbage in any workflow. Make a few quick edits so it fits your actual workflow even better.

[K T Tzy. @KTTzyyy]

@ollama Why the hell you remove Minimax M3?

The limits are also horrible

[Su @Sukiea1008]

# AGENTS.md

## 目的
用最小充分方案完成当前任务。
禁止过度工程化。
规划可以偏强，执行必须偏轻。
不能证明必要的设计，默认不做。
不能证明必要的测试，默认不加。

## 工作流
1. 先理解需求，再动手。不要先改代码再猜意图。
2. 规划阶段可用较高推理。执行阶段默认用中低推理，或改用更轻量模型落地。
3. 不要全程开启最高推理档位。
4. 不要默认并行拉起多个 Agent。一个任务先单线做完，再决定要不要拆。
5. 只启用完成任务所必需的 skill。不要安装重流程 skill。
6. 先产出最小计划，再执行。计划必须写清：
- 目标
- 非目标
- 验收标准
- 不改动的范围

## 失败模式
1. 没有真正理解意图，只修了表面问题。
2. 本可以做一次干净的根因修复，却用历史补丁、兼容层、双轨实现、副本和分支把代码堆胖。
3. 为很少发生的情况过度设计，让日常维护成本变高。
4. 判断依据错了，推理再完整结论也是错的。
5. 本该直接读代码定位问题，却用检索或猜测替代阅读。
6. 把“补测试”当成继续加抽象、扩范围、显得完整的借口。

## 行动边界
1. 动手前先复述：
- 用户真正想要什么
- 本次范围是什么
- 明确不做的事
- 怎样算完成
2. 任何不可逆操作，必须等待用户回复确认口令后再执行。
- 确认口令由用户指定。
- 没有口令、口令错误、或其他回复，一律拒绝执行。
3. 以下操作默认不算不可逆，可以执行：
- Git 回滚、还原、切分支
- 把文件移动到当前仓库的备份目录
- 跑测试、查看 diff、生成计划、只读分析
4. 发现自己正在做以下事情时，必须停下来改成更小方案：
- 新增抽象、框架或配置层，但当前需求并不需要
- 为了以后可能用到而提前设计
- 为了满足约束而继续叠加更多约束
- 同时改很多无关文件
- 创造第二套实现来兼容旧逻辑
- 借机补一套完整测试体系

## 测试
测试只服务于当前改动的验收。
测试不负责补齐历史覆盖率，不负责设计未来测试体系。

1. 优先跑与本次改动相关的现有测试。
2. 现有测试能证明改动正确，就不要新增测试。
3. 只有在下面两种情况才允许新增测试：
- 这次改了行为，但现有测试盖不到这个行为
- 用户明确要求补测试
4. 新增测试最多覆盖本次真正改动的 1 个主路径，必要时再加 1 个关键失败路径。
5. 禁止为了更完整而扩大测试范围。
6. 禁止借机补全无关模块的测试。
7. 禁止引入新的测试框架、测试工具或测试基础设施。
8. 禁止写大量快照、参数化矩阵或端到端套件。
9. 禁止为当前需求未要求的边界写测试。
10. 禁止先改测试再倒逼产品行为变复杂。
11. 禁止把测试变绿当成可以继续加抽象的理由。

新增任何测试前，必须能回答：
- 这个测试在验证哪个已被接受的需求
- 去掉它，现有测试是否就无法发现这次回归
- 它是不是比实现本身更复杂

如果测试代码比实现代码更长或更绕，默认视为过度工程，应删测试或缩小实现。

## 模型分工
- 需求澄清和方案审查：用较强模型
- 写代码、改代码、跑测试：用中低配模型，或更轻量的执行模型
- 发现执行模型开始叠架构、加兼容、扩范围、补大套测试时：立刻停下来，重写最小计划

## 完成前检查
- 已复述意图和验收标准
- 方案是最小方案，不是最大方案
- 已标明非目标
- 优先读了相关代码，而不是靠检索拼结论
- 只改了完成任务所需的最小文件集
- 相关现有测试已跑过
- 没有为未要求的场景新增测试
- 若新增测试，只锁本次行为，且条数很少
- 测试没有引入新依赖或新目录结构
- diff 小，无多余文件，无残留调试代码
- 没有为了看起来完整而额外施工

## 总则
先确认意图，再用最小改动完成验收。
设计不能证明必要，默认不做。
测试不能证明必要，默认不加。

[Andrew Ng @AndrewYNg]

How have software engineering fundamentals changed with agentic coding? Here is our AI Engineering Skills map for software engineering fundamentals. https://t.co/cnRLj43DLs

[Vox @Voxyz_ai]

If I could give Codex / Grok Bot just one skill for drawing system diagrams, I’d pick 𝗔𝗿𝗰𝗵𝗶𝗳𝘆.

It already has 29K+ stars on GitHub.

I now use it by default whenever I need to map a system.

Give it the code, then use this template:

“Turn [CODEBASE / FEATURE / WORKFLOW] into a [DIAGRAM TYPE] diagram. Choose from Architecture, Workflow, Sequence, Data Flow, or Lifecycle.”

Once installed, your agent can:

→ Create one self-contained HTML file and export it as PNG, SVG, or video
→ Search for any component and trace its upstream and downstream paths
→ Compare before and after versions side by side to see what was added, removed, or rearranged

𝗜𝗻𝘀𝘁𝗮𝗹𝗹 𝗔𝗿𝗰𝗵𝗶𝗳𝘆 𝗴𝗹𝗼𝗯𝗮𝗹𝗹𝘆:

npx skills add tt-a1i/archify -g

Then read this repo and use Archify to generate a system diagram.

Choose the diagram type that best fits what I need to understand: Architecture, Workflow, Sequence, Data Flow, or Lifecycle.

Keep the 8–12 most important components and one primary path. Label external services, databases, and component connections clearly. Put supporting details in cards instead of adding more edges to the main diagram.

Generate and verify a standalone HTML file. Confirm that component search and upstream/downstream tracing work, then export a high-resolution 2400 × 1260 PNG.

𝗢𝗽𝗲𝗻 𝘀𝗼𝘂𝗿𝗰𝗲 𝗿𝗲𝗽𝗼 👇
https://t.co/HdNaUgt1jx

[Pranav @pranavcmadhukar]

Solved! If you want Codex to get better at building harnesses/ agents, just download this book and tell it to read it, specifically the parts on context engineering and agent "eyes".

https://t.co/rpn1qAaLIo

(I have no idea who the author is and am not affiliated)

[K T Tzy. @KTTzyyy]

@direkturcrypto Fakta bahwa di Tiktok masih banyak yang begini is crazy

Gw aja udah ninggalin ChatGPT, Gemini web, https://t.co/n1H3T3xpX9 web sejak ada Hermes (apalagi Hermes Desktop), ZCode, Codex GUI, dan harness-harness lain

[Ben Holmes @BHolmesDev]

Introducing /skill-doctor v1

Skill doctor is a skill that improves your skills:

- Crawls your past CC, Codex, or Warp convos
- Scores them for efficiency and code quality
- Proposes diffs to your skills that you'd actually merge (STE-100 approved)

https://t.co/vU0OeF1S2c 🔖

[Paidax @xin_pai88825]

想要 Vibecoding 出来的界面美观，那就不要错过下面几个组件库

1. beUI ：页面总差一点“活起来”的感觉？逛逛 beUI，它的动效做得很顺，复制进项目就能用。

2. BoardUI：如果你在做后台、数据看板或者 Agent 产品，BoardUI 能帮你省掉很多从零搭界面的时间。

3. ThreeUI：想给网站加点 3D、WebGL 效果，又不想自己从头啃 Three.js，可以直接来这里找现成的。

4. Fluid Functionalism：这个站挺有意思，不做满屏乱飞的特效，而是认真研究怎么用动画让界面变得更好用。

5. Watermelon UI：里面有不少能直接放进真实项目的组件和完整页面，很适合边逛边找灵感，一不小心就会看很久。

6. HeroUI Pro：如果你想要一套能把设计、前端和 AI 工具串起来的 UI 系统，HeroUI Pro 做得相当完整。

[Ram Maheshwari @rammcodes]

WTF Broooooooo???? .... This feels like a goldmine for vibe coders 🌋

It's a collection of free UI prompts for things like hero sections, pricing pages, auth screens, bento grids, dashboards, and more.

You can copy a prompt and give it to your AI coding agent to generate better-looking, less generic UI without having to describe every design detail yourself.

Perfect if you build websites with AI and want your UIs to stand out :)

Follow me for more amazing Coding, AI & Web Development insights 💎

[Eric Ciarla (hiring) @ericciarla]

introducing firecrawl free keyless

now your ai agents can search and scrape the web 100% free

- state-of-the-art search accuracy (94.7% on simpleqa)
- sub-3s scrapes, any page into clean markdown
- no api key, no signup

npx -y firecrawl-cli@latest init --all --browser