tobwen

OpenRouter may have leaked something interesting: UKExpertMedical, a UK medico-legal agency, is using Google Flash AI. Let's hope it's not for personal medical data or to draft medico-legal reports. At least "Enhanced Medical Document Analysis" sounds like it.

Original URL: https://openrouter.ai/apps?url=https%3A%2F%2Fgithub.com%2FUKExpertMedical%2Ffixit-blank-pages

Sadly, the repository has gone private: https://github.com/UKExpertMedical/fixit-blank-pages

Over the last few days, tons and tons of `SKILL.md` files have been popping up on marketplaces everywhere.

There are tons of tutorials on Youtube, how to use Claude to create a skill out of nothing. I bet no human ever reviewed, or optimized those texts. So instead of feeding the AI the real experience and skills of a human coder, we're stuffing it with input based on what the AI thinks a human would do.

Does this actually lead to better code, or is it just a vicious loop of "SISO" (shit in, shit out)?

There are many CLI AI agent tools out there these days: aider, droid, codex, copilot, crush, opencode, etc. Some are "locked" to major providers and subscription-only, some play nice with BYOK or multi-provider proxies like OpenRouter, and others can be "made" to support multiple backends (EULA violations, etc.).

Is there an "awesome CLI AI agents" repo or site curating them? And maybe some solid reviews or head-to-head comparisons floating around?

I've tested most of them on my own and there are significant differences. I'd love to dump my findings into a post, but I'm not sure if it'd hold a candle to those big, polished code benchmarks out there.