First world problems with AI agents

People want AI at work — not at play. Social networks should double down on anti-bot measures.

If your AGENTS/ CLAUDE md file doesn’t include linting rules, you’re doing it wrong. Telling your AI agent to continuously lint the code it generates is probably the single most important thing you can do to increase productivity when vibe-coding/vibe-engineering.

If you have a type-check rule in your AGENTS/CLAUDE md file, check out TypeScript Go and run `tsgo` instead of `tsc`:

Is there any EIP-7702 registry of universal batch contracts? Ideally deployed at the same deterministic address across chains.

Is there any way enable email notifications for Take Profit/Stop Loss events on Hyperliquid?

One of AI’s killer use cases is using AI to improve itself. e.g. having an LLM refine your prompt before you feed it back into the same LLM.

This is cool - a tool for jailbreaking Claude Code. With tweak cc, you can customize anything and everything, including the system prompts and what tools Claude knows about.

many MCPs are just glorified wrappers around `cli-name --help`

Anthropic's plugin marketplace for Claude Code is a treasure trove. You should really take a look at this if you're using Claude Code:

Anthropic should dial back Claude’s obsession with writing documentation for the code it generates. Opus 4.5 is better than previous models, but it still over-documents (especially in files like README). By default, the model should write for humans — not for other AIs.

Opus 4.5 is the most impressive model I've ever used. It's incredibly intelligent.

Opus 4.5 is the first model that does well even after compacting. Opus 4.1, Sonnet 4.1, and Sonnet 4.5 were all pretty bad at this stage. I was always starting a new chat.