"If you only do what you can do, you'll never be better than what you are"

Love code review

Kimi seems good too on this test.

Claude sonnet 4.5 correct user with "you mean using fzf as a menu with actions! Yes, fzf supports this with --bind for custom key actions:", This is good and correct output.

Testing models's hallucination, minimax + glm 4.7 done bad.

Done some improvement on the explore's prompt, it did significantly improve the search accuracy and speed.

I like this diff tool, with review comments. https://t.co/xD5fpqTopS #vibecoding #code_review

我说咋好久没看到这位老哥的推了，原来把我 block了。没有印象为啥。

Agents communicate with each other by tmux

I want the main agent plan for the doc, then spawn another rush agent to implement it.

spawn tmux panel to run agent works.

This auto review workflow is sooo good!

Good review

automatically review with another model after a few turns when there are code changes. The review request has an interactive UI for confirmation. #PI

Create a deepwiki tool in #pi agent

Don't know what is wrong with the amp smart. Sorry for my bad language.

@mitsuhiko doing the same with a few react quirks the models love but using biome's grit plugins https://t.co/jXBsKKiTkp

recent performance

My 2025 OpenRouter Wrapped is here! 🏆 Top 2% of MiniMax M2 users. #OpenRouterWrapped

The reason people like long context chat is because the compact/compress taking too long, like it did in claude/factory-droid

after "/new", the agent can *directly* know what I am talking about.