nft//defi//OG⏰ 🔮 (@opinion_bits) profile picture

nft//defi//OG⏰ 🔮

@opinion_bits

Founder x2, VC-backed once. A decade shipping software, now shipping with agents. Building https://nitter.cf/t.co/kb9pjtLdKW , https://nitter.cf/t.co/UADMeORlfw , @GrinderWiki

 BuilderSphere
Joined April 2019
  • Tweets2.5K
  • Following535
  • Followers667
249Media
Photo by nft//defi//OG⏰ 🔮 (@opinion_bits)Photo by nft//defi//OG⏰ 🔮 (@opinion_bits)Photo by nft//defi//OG⏰ 🔮 (@opinion_bits)Photo by nft//defi//OG⏰ 🔮 (@opinion_bits)Photo by nft//defi//OG⏰ 🔮 (@opinion_bits)Photo by nft//defi//OG⏰ 🔮 (@opinion_bits)Photo by nft//defi//OG⏰ 🔮 (@opinion_bits)Photo by nft//defi//OG⏰ 🔮 (@opinion_bits)Photo by nft//defi//OG⏰ 🔮 (@opinion_bits)Photo by nft//defi//OG⏰ 🔮 (@opinion_bits)
Claude Code Plugins 99% of Devs Have NEVER Seen, My Tier List Will Change How You Code Forever

Claude Code Plugins 99% of Devs Have NEVER Seen, My Tier List Will Change How You Code Forever

Article · @opinion_bits
1
1
17
6,481
Web scraping is getting harder. JS rendering. DOM changes. Bot checks. Lazy loading. Huge pages with tiny useful content. 8 OSS repos worth knowing: Firecrawl — 198K★ github.com/firecrawl/firecra… Scrapling — 95K+★ github.com/D4Vinci/Scrapling Crawl4AI — 86.8K★ github.com/unclecode/crawl4a… Scrapy — 64.6K★ github.com/scrapy/scrapy ScrapeGraphAI — 32.3K★ github.com/ScrapeGraphAI/Scr… Crawlee — 25.8K★ github.com/apify/crawlee Camoufox — 12.4K★ github.com/daijro/camoufox Trafilatura — 6.9K★ github.com/adbar/trafilatura The new stack: URL → browser/HTTP → survive DOM changes → extract → clean → JSON/Markdown → agent Best rule: Use the cheapest deterministic layer that works. Escalate only when the web fights back.
1
1
26
7 OSS repos that show exactly where your AI app might be quietly burning money: RagaAI Catalyst — 16.2K★ github.com/raga-ai-hub/RagaA… Evidently — 8K★ github.com/evidentlyai/evide… OpenLLMetry — 7.5K★ github.com/traceloop/openllm… Helicone — 6.3K★ github.com/Helicone/helicone AgentOps — 5.9K★ github.com/AgentOps-AI/agent… Langfuse — 36.9K★ github.com/langfuse/langfuse Phoenix — 12K★ github.com/Arize-ai/phoenix The real production trace is: request → model → tool → retry → subagent → tokens → latency → cost Traditional logs tell you something failed. LLM observability tells you the 37 decisions it made before failing.
1
18
DeepSeek just open-sourced a GitHub superagent that can work through issues + PRs with multiple agents in parallel. And the rest of its OSS stack is just as interesting: dsh-superagent-prototype — 10.9K★ github.com/deepseek-ai/dsh-s… DeepInfer — 24.9K★ github.com/deepseek-ai/DeepI… FlashMLA — 13.2K★ github.com/deepseek-ai/Flash… DeepEP — 10.4K★ github.com/deepseek-ai/DeepE… DeepGEMM — 9.1K★ github.com/deepseek-ai/DeepG… 3FS — 10.3K★ github.com/deepseek-ai/3FS The stack: GitHub issue → autonomous worker → inference → attention → GPU compute → storage DeepSeek is not just open-sourcing models anymore. It is open-sourcing the machine around the model.
1
48
AWS just open-sourced Strands Harness and says it can cut agent costs vs Claude Code/Codex while keeping broadly comparable accuracy. The bigger story: the harness matters almost as much as the model. 6 OSS repos worth watching: Strands Harness — 9K★ github.com/strands-agents/ha… DeepSeek Harness — 267K★ github.com/deepseek-ai/deeps… OpenCode — 223K★ github.com/anomalyco/opencod… OpenHands — 92K★ github.com/OpenHands/OpenHan… goose — 56K★ github.com/aaif-goose/goose Aider — 49.9K★ github.com/Aider-AI/aider The real stack is: model + context + tools + caching + memory + verification = agent Sometimes the cheapest upgrade is not a cheaper model. It is a better harness.
53
Andrej Karpathy quietly built one of the best free AI curricula on GitHub. 7 repos. 300K+ stars. Follow them in this order: micrograd — 18.3K★ github.com/karpathy/microgra… nn-zero-to-hero — 20.3K★ github.com/karpathy/nn-zero-… minbpe — 10.8K★ github.com/karpathy/minbpe nanoGPT — 64.3K★ github.com/karpathy/nanoGPT llm.c — 31.3K★ github.com/karpathy/llm.c nanochat — 59.1K★ github.com/karpathy/nanochat autoresearch — 97.8K★ github.com/karpathy/autorese… The path: backprop → neural nets → tokenization → GPT → CUDA → full chat model → autonomous research Karpathy basically takes you from “LLMs are magic” to “I understand the stack well enough to automate experiments.”
29
Claude Code Mods are getting weirdly powerful. They can intercept tool calls, manage context, redraw the UI + change what happens around the agent. 6 OSS repos worth watching: Firstmate — 8.4K★ github.com/kunchenguid/first… fast-jev-compaction — 7.8K★ github.com/tamaratran/fast-j… terminal-browser — 4.3K★ github.com/zenbu-labs/termin… Agent Time Machine — 1.1K★ github.com/YeonwooSung/my-cl… TextFold — 800+★ github.com/chenxiachan/textf… Context Attachment Manager — 590+★ github.com/chipfighter/conte… Skills change what Claude knows how to do. MCP changes what it can reach. Mods change what happens around the agent while it works.
1
1
70
Hermes is quietly becoming an agent runtime, not just another model. 6 OSS repos building around it: Paperclip — 108K★ github.com/paperclipai/paper… Open WebUI — 154K★ github.com/open-webui/open-w… Buzz — 36.6K★ github.com/block/buzz hermes-agent-self-evolution — 5.6K★ github.com/NousResearch/herm… Mnemosyne — 3.3K★ github.com/mnemosyne-oss/mne… SkillClaw — 2.7K★ github.com/AMAP-ML/SkillClaw The stack is becoming: model → Hermes runtime → memory + skills → tools → workspace → specialist agent The model is becoming replaceable. The persistent system around it is becoming the product.
174
RIP the $100K document extraction stack. 7 OSS repos now cover almost the whole pipeline: LangExtract — 39K★ github.com/google/langextrac… PaddleOCR — 91K★ github.com/PaddlePaddle/Padd… MinerU — 81K★ github.com/opendatalab/Miner… Docling — 69K★ github.com/docling-project/d… Marker — 40K★ github.com/datalab-to/marker Unstructured — 15K★ github.com/Unstructured-IO/u… Instructor — 14K★ github.com/567-labs/instruct… The stack is becoming: PDF/image → OCR → layout → structured doc → extraction → validation → source evidence The moat is shifting from owning the extractor to assembling the pipeline better.
3
83
AI trading should not mean “ask an LLM if NVDA goes up.” 9 OSS repos building the actual stack: TradingAgents — 113K★ github.com/TauricResearch/Tr… ai-hedge-fund — 64K★ github.com/virattt/ai-hedge-… Freqtrade — 55.8K★ github.com/freqtrade/freqtra… Qlib — 49.7K★ github.com/microsoft/qlib FinGPT — 21.5K★ github.com/AI4Finance-Founda… RD-Agent — 15.1K★ github.com/microsoft/RD-Agen… FinRobot — 8.3K★ github.com/AI4Finance-Founda… FinRL-Trading — 3.9K★ github.com/AI4Finance-Founda… LumiBot — 2.2K★ github.com/Lumiwealth/lumibo… The better stack: research → signals → debate → portfolio → risk → backtest → execution
41
RAG is getting smaller, local + way more interesting. 8 OSS repos rethinking retrieval: LEANN — 13K★ github.com/StarTrail-org/LEA… PageIndex — 38.9K★ github.com/VectifyAI/PageInd… Khoj — 37.8K★ github.com/khoj-ai/khoj Memvid — 16.6K★ github.com/memvid/memvid txtai — 13K★ github.com/neuml/txtai LanceDB — 11.8K★ github.com/lancedb/lancedb sqlite-vec — 8.1K★ github.com/asg017/sqlite-vec USearch — 4.4K★ github.com/unum-cloud/USearc… The old stack was: chunk everything → embed everything → store everything The new question is: how little infrastructure can retrieval actually need?
56
GitHub is full of projects that feel like they leaked from 2030. 10 worth exploring: PhotoCraft — 72.6K★ github.com/storytold/photocr… Magpie — 20.8K★ github.com/yetone/magpie Anatomy — 20.2K★ github.com/wheresryan22/anat… Universal Modder — 8.9K★ github.com/rehan-remade/univ… Edge0 — 6.8K★ github.com/Edge0-AI/Edge0 Coucou — 6.2K★ github.com/Louis-CFM/coucou Replica Skill — 5.3K★ github.com/Jakeschincariol/r… Human Atlas — 4.2K★ github.com/ashemag/human-atl… Shaders — 2.4K★ github.com/shader-effects-in… Mailcheap — 578★ github.com/marclou/mailcheap Photoshop in Rust. AI game modding. Models streamed from SSD. Interactive anatomy. Agent control panels. GitHub is still where people build things you didn’t know were a category yet.
101
AI made answers free. These OSS repos are trying to make AI actually teach you: DeepTutor — 42.3K★ github.com/HKUDS/DeepTutor learn — 3.4K★ github.com/amosblomqvist/lea… LearnHarness — 310★ github.com/amosblomqvist/lea… OpenTutor — 149★ github.com/zijinz456/OpenTut… Tutor MCP — 45★ github.com/ArnaudGuiovanna/t… TuTor — 6★ github.com/kevinnio/tutor The better learning loop: diagnose → explain → make you try → test → remember the gap → test again later An AI tutor should make itself less necessary over time.
1
77
Don’t start your next Opus 5.5 video from scratch. 8 OSS repos with 33K+ combined stars - finished examples, prompts, styles + source: video-shotcraft — 12K★ github.com/Vincentwei1021/vi… awesome-opus5-5-videos — 9K★ github.com/yihui-dev/awesome… lemo-opuscar — 5K★ github.com/lemomo-ai/lemo-op… pdoom-video — 3.2K★ github.com/mexicat/pdoom-vid… mg-styles-15 — 1.7K★ github.com/Vincentwei1021/mg… ClaudeAnimationBase — 1.5K★ github.com/JohnHeibel/Claude… awesome-claude-5-5-videos — 600★ github.com/athemeroy/awesome… opus-video-skills — 250★ github.com/tuzhechen2005/opu… reference → steal the motion language → fork working code → let Opus remix it
1
1
156
You upgraded your model. Your SKILL.md is still living three releases ago. The skill that was relevant for GPT-5 may add unnecessary work for GPT-6. Same with Claude Code Models. I built Skill Adapter to help old skills adapt without losing the workflow you liked. Original untouched. Changes explained. MIT licensed. github.com/smukh/skill-adapt…
GitHub - smukh/skill-adapter: Adapt existing agent skills to your model while preserving their...

GitHub - smukh/skill-adapter: Adapt existing agent skills to your model while preserving their...

Adapt existing agent skills to your model while preserving their purpose, workflows, and requirements. - smukh/skill-adapter

github.com
3
1
4
118
nft//defi//OG⏰ 🔮 retweeted
GitTrend (@GitTrend0x) profile picture
GitTrend
@GitTrend0xSep 14
人人推荐的 Skills 🔥 1️⃣ anthropics/skills（github.com/anthropics/skills） Anthropic 官方公开仓：docx / pdf / pptx / xlsx 生产文档技能 + 模板 + skill-creator。技能格式的出厂说明书。 npx skills add anthropics/skills “技能第一次从协议制定方的仓库里长出来，而不是社区口口相传的咒语”！ 2️⃣ addyosmani/agent-skills（github.com/addyosmani/agent-…） 24 个生命周期技能 + 9 条斜杠入口：/spec → /plan → /build → /test → /review → /ship，每步带验证门和反借口表。 npx skills add addyosmani/agent-skills “技能第一次按发版流水线出货，而不是再塞一条万能 prompt”！ 3️⃣ nextlevelbuilder/ui-ux-pro-max-skill（github.com/nextlevelbuilder/…） 跨端 UI/UX 设计智能：设计系统、推荐、多平台视觉宪法。安装榜长期前排，12 万+ 星。 npm i -g ui-ux-pro-max-cli “界面第一次有可执行的审美，而不是『做漂亮一点』四个字”！ 4️⃣ NVIDIA/SkillSpector（github.com/NVIDIA/SkillSpect…） 装技能之前先安检：提示注入、数据外带、供应链、恶意模式，静态分析 + 可选语义评估。 uv tool install git+github.com/NVIDIA/skillspect… “技能战争进入安检门时代。能装和敢装，不是一回事”！ 5️⃣ Neeeophytee/finding-unknowns-skills（github.com/Neeeophytee/findi…） 13 个「先找出你不知道自己不知道什么」的技能：盲区巡检、假设拷问、访谈、引用狩猎、合并前小测验。Claude / Codex / Hermes 通吃。 npx skills add Neeeophytee/finding-unknowns-skills “需求第一次在开工前被盘问到能执行，而不是上线当天才发现少问了那一句”！ // 总结 2026 的技能战争已经从「再写一条 prompt」变成「官方出厂源 + 发版流水线 + 能进稿的设计宪法 + 安装前安检 + 未知项狩猎」。谁先把这五件套装进自己的 Agent，谁就把下一波生产力锁死在自己终端里。直接冲。
GitHub - anthropics/skills: Public repository for Agent Skills

GitHub - anthropics/skills: Public repository for Agent Skills

Public repository for Agent Skills. Contribute to anthropics/skills development by creating an account on GitHub.

github.com
GitTrend (@GitTrend0x) profile pictureGitTrend
@GitTrend0xSep 13
人人都推荐的 Skills 🔥 1️⃣ vercel-labs/agent-skills （github.com/vercel-labs/agent…） Vercel 官方技能仓：find-skills / agent-browser / frontend-design / react 最佳实践，skills.sh 安装榜长期第一。 npx skills add vercel-labs/agent-skills “技能货架第一次从平台厂出厂，而不是社区口口相传的咒语”！ 2️⃣ alirezarezvani/claude-skills（github.com/alirezarezvani/cl…） 380+ 技能 / 30+ Agent / 70+ 命令，通吃 Claude Code、Codex、Gemini、Hermes、Cursor 等 13 家 harness。 “技能第一次按公司编制出货：工程、增长、合规、C-level 顾问一仓装完”！ 3️⃣ adityaarakeri/senior-agent-skills（github.com/adityaarakeri/sen…） 十本可移植作战手册：repo-recon → plan-first → tdd-loop → debug-protocol → safe-refactor → self-review → verify-done。 “技能第一次逼 Agent 像谨慎的高工办事，而不是热情实习生到处撒补丁”！ 4️⃣ K-Dense-AI/scientific-agent-skills（github.com/K-Dense-AI/scient…） 165 个过验证的科学技能：生物、化学、医学、药物发现。不是又一篇「帮我写论文」prompt。 “实验室工作流第一次有可检查的技能宪法，而不是聊天记录里的愿望清单”！ 5️⃣ mattpocock/skills（github.com/mattpocock/skills） grill-me / grill-with-docs：先把需求烤清楚再动手。安装榜和「值得装」榜双杀。 “需求第一次被技能盘问到能执行，而不是一句『做个登录页』就开工”！ // 总结 2026 的技能战争已经从「再写一条 prompt」变成「平台厂官方源 + 跨 harness 军团包 + 高工作战手册 + 科学验证集 + 需求拷问术」。谁先把这五件套装进自己的 Agent，谁就把下一波生产力锁死在自己终端里。直接冲。
GitHub - vercel-labs/agent-skills: Vercel's official collection of agent skills

GitHub - vercel-labs/agent-skills: Vercel's official collection of agent skills

Vercel's official collection of agent skills. Contribute to vercel-labs/agent-skills development by creating an account on GitHub.

github.com
1
2
1
9
5,185
Built Roll Credits, a Claude Mod that gives your coding session movie credits. You direct. Your most-edited file stars. Failed tool calls become plot twists. Runs locally. No extra model calls. MIT licensed. Early access. github.com/smukh/roll-credit…
🤖 Made with AI
Image from nft//defi//OG⏰ 🔮 (@opinion_bits)
1
3
67
Most AI coding agents can add an ARIA attribute. Far fewer can prove the original accessibility issue is actually fixed. We built a11y-agent-skills, a portable set of 7 open-source skills for: • Accessibility audits • Source-level fixes • Forms and dialogs • Component reviews • Keyboard navigation • Regression tests Your agent investigates and edits the code. Playwright and axe-core produce deterministic evidence, stable fingerprints, and verification reports. It works across Codex, Claude Code, Cursor, and other Agent Skills-compatible hosts. It also ships as an npm CLI, MCP server, and GitHub Action. Local-first. No telemetry. No API key. Repo in first comment 👇
4
1
6
163
Github Repo: github.com/smukh/a11y-agent-…
GitHub - smukh/a11y-agent-skills: Tested accessibility skills, CLI, MCP server, and GitHub Action...

GitHub - smukh/a11y-agent-skills: Tested accessibility skills, CLI, MCP server, and GitHub Action...

Tested accessibility skills, CLI, MCP server, and GitHub Action for agent-assisted audits and regression checks. - smukh/a11y-agent-skills

github.com
3
63
flowstacks.xyz it's free??? how? what a tool
FlowStacks — Verified AI workflows that actually ship

FlowStacks — Verified AI workflows that actually ship

A structured, searchable library of verified AI workflows from builders. Browse by tool, by use case, or by what they replace, every recipe shows its stack, its cost, and how it was verified.

flowstacks.xyz
1
59
AI doesn’t need another humanizer. It needs stylebooks. Now it covers 16 installable writing styles based on publicly available rule sets from organizations like Google, ukgov, Microsoft etc. github.com/Neeeophytee/agent…
GitHub - Neeeophytee/agent-stylebooks: 11 installable editorial systems for AI agents, based on...

GitHub - Neeeophytee/agent-stylebooks: 11 installable editorial systems for AI agents, based on...

11 installable editorial systems for AI agents, based on leading public style guides. - Neeeophytee/agent-stylebooks

github.com
4
60
Load more