Claude Code grows up, Cloudflare ships agent skills

DEEP DIVE

Claude Code Guide 2026: 25 Features with Examples — a sprawling reference covering CLAUDE.md, skills, subagents, hooks, MCP and Auto Mode. The framing — "Claude Code is a layered agentic coding tool, not a single chat prompt" — is the right mental model if you're still treating it like a glorified REPL. Worth a slow read before your next agent architecture decision. — Michal Sutter

Cloudflare One stack: agent-powered deployment — a library of agent skills that "gives any AI agent the knowledge it needs to plan, deploy, and manage a Zero Trust environment." Translation: less prompt engineering, more vendor-shaped scaffolding. Useful if you're already on Cloudflare; another lock-in vector if you're not. — blog.cloudflare.com

Coinbase for Agents: automating portfolio trading with AI — lets agents execute trades and payments directly against user portfolios — closing the gap Coinbase describes as LLMs that "process vast quantities of data but lack direct integration with active financial portfolios." The custody and prompt-injection surface here deserves more scrutiny than the launch post offers. — Ryan Daws

4 Lines You Should Include in Your Claude Skill — a short, opinionated checklist of four boilerplate lines that meaningfully reduce hallucinations in Claude Skill definitions. The kind of micro-pattern that quietly separates demoware from things you'd actually ship to users. — Haden Pelletier

Slightly reducing the sloppiness of AI-generated front end — practical, unglamorous techniques for cleaning up the visual and structural artifacts that LLMs habitually produce in UI code. "Slightly" is doing honest work in that title, and that's why it's worth reading. — envs.net

POLICY

The Anthropic models ban was never about a jailbreak — reporting suggests the US government restriction on Anthropic models was driven by policy friction, not a security incident. If you're modeling regulatory risk for your stack, assume geopolitics, not CVEs. — amarant

My Homelab AI Dev Platform — a detailed write-up of running open models locally for dev work. Single-machine constraints are real, but so is the API bill. — rsgm

Paca — lightweight Jira alternative for human-AI collaboration — open-source task tracker designed assuming agents are first-class contributors. Whether that's a genuine new category or a Jira plugin in disguise is still an open question. — pikann22

Launch HN: Adam (YC W25) — Open-Source AI CAD — an open-source attempt at LLM-driven parametric CAD. Interesting space; the bar set by Fusion 360 plugins and generative incumbents is high. — zachdive

Accenture: 74% would trust an AI agent over their best friend to shop — a 25,590-person survey across 16 countries finds "74% of respondents would trust a personal AI agent more than their best friend" for purchase decisions. Stated-preference data, so discount accordingly — but useful ammo if you're pitching an agent-commerce deck. — Muhammad Zulhusni