I have test automations that run through a large set of popular open models, many require careful harness engineering to work well.
@steipete
-
Automated PR creation from issues matching VISION.md
By
–
Whenever you create an issue on one of oure open source projects, @clawsweeper will review it, and *if* it fits the VISION.md file, will pick it up and create+autoreview a PR. e.g.:
-
Coding intelligence reduces bugs; ds4 engineering is great
By
–
IMO for coding you want ad much intelligence as possible to ship better code with fewer bugs. ds4 is great engineering tho.
-
Comparing token count to coding, can use Sonnet-class model
By
–
compared to coding these are very few tokens, and you can even use a Sonnet-class model (= gpt 5 mini)
-
It runs on Codex, Copilot, or their own harness
By
–
it runs on codex now. or copilot. or our own harness!
-
User questions if ‘load-bearing’ is unique to Claude, not Codex
By
–
is load-bearing a Claude thing? Haven’t seen that on Codex. Maybe I don’t have enough load
-
Running prompts from a project folder containing all projects
By
–
I run most prompts of a project folder all my projects are in, works fine?
-
Prompting ‘do whatever you need to do’ yields exactly that
By
–
If you prompt "do whatever you need to do" you'll get exactly that.
-
Subscription cost comparison: Chinese labs cheaper but older quality
By
–
The 100/200$/month sub works well for most. There’s also really competitive offering from chinese labs out, quality is usually a year behind latest gen but for 10% of the costs.
-
Cloud agents eliminate need for open MacBooks
By
–
Fewer people having to run around with their MacBooks open as it makes running agents in the cloud easy.