Hacker News Reader

moofeezon AI coding has made CI a bottleneck, so we reworked ours to keep up

We actually tried this and (for us) it doesn't lead to enough of a gain that's worth the extra complexity.

moofeezon Post-trained Qwen3-Coder with a debugger: 70% → 89% solve rate, 59% fewer turns

I post-trained Qwen3-Coder to effectively use a real debugger harness (stepping through code, inspecting runtime state, and reasoning from execution traces) instead of reasoning only over the static code.

Results on 27 held-out bug tasks:

- Base model + debugger harness (no RL): 67% solve rate, only ~5% fewer turns

- After post-training with RL: 70% → 89% solve rate, 46 → 19 median turns (−59%)

The key was teaching the model how to use the debugger effectively. Just giving the base model the harness did not help.

More detailed results and a full blog post coming soon. Happy to answer questions in the meantime!

moofeezon Show HN: Conductor, a Mac app that lets you run a bunch of Claude Codes at once

https://github.com/smtg-ai/claude-squad sounds like it might a better fit for your usecase, free and open-source :)

moofeezon Show HN: Conductor, a Mac app that lets you run a bunch of Claude Codes at once

hey, co-creator of Claude Squad – imo the most popular+used of these "claude code multiplexers" and it's also open source and free :)

works like you'd expect

https://github.com/smtg-ai/claude-squad

moofeezon Ask HN: Airdrop alternative that works on all platforms?

Yesterday, I posted about a cross-platform AirDrop alternative that I'm working on: https://x.com/moofeez/status/1744119063901544530?s=20

Would love to hear your feedback on what you'd like to see!