I'm asking the question focusing specifically on what I could use/download/configure right now, not necessarily on any custom written harness you might be using which isn't shared and fits your specific workflow.
I have been finding Alibaba's OCR (https://github.com/alibaba/open-code-review) very useful and getting good results as part of my own automated flow and at work. However, apart from this project, I haven't found anything else out there that gives me some deterministic output so that even when running multiple LLMs on various effort levels I get mostly the same 2-3 high-importance fixes.
Jev seemed to be quite a big change and I've noticed some people using it as a decision model alongside a request router - jev decides where a request goes and aims to save you money, but even that doesn't seem to hit Hacker News that much, so apart from the few medium articles, I haven't seen much more around that.
People in the comments seem to build their own harnesses, discuss their own specific tools that might be vibecoded or not (I don't mean this in a derogatory way, but rather state it as fact - I have an easier time navigating non-LLM code for the most part).
However, I don't seem to find a lot of tooling that can be general purpose and added into an existing harness via a skill and have it be invoked as such or as an MCP/plugin/whatever else. Not that it makes it to this page at least.
So to repeat the question - is there anything that you use that is available out there that helps your workflow, gets you better results in terms of lowering resource usage, costs and/or quality?
On the other hand, if there really isn't anything out there or it's slim pickings, what made you build your own tool, how did you start the process and how have you measured its impact upon the previously mentioned metrics?
Comments URL: https://news.ycombinator.com/item?id=50439324
Points: 1
# Comments: 1
