The ADHD project for Claude has reached over 500 GitHub stars and is being integrated into major open-source projects. The developer announces plans to make it compatible with local models, create a standalone interface, run large-scale benchmarks, and open community registrations.
Hey community, what's up? First of all, congratulations because, other than getting featured on top tech media and hitting more than 500 GitHub stars, we are now also a major part of tens of major open source projects who are now using us for their code reasoning module. But hey, one question remains: where does this head from now? If you guys hadn't given it such a super response, I would have said that okay, we will now publish in some good tech journal or some neuro or AI journal, but now things are different. So I have a couple of updates about what I am planning next, and I want you to give me feedback on what you think about it. 1. **Making ADHD global OSS compatible** \- so right now, it is a skill for Claude, but that is taking a lot of quota and limits from most of the people who are trying it, so I'm planning to make it compatible with all the local models, Openclaw, and all other providers as well. (By the way, the Openclaw community on Reddit has already adopted our ADHD skill as their default). 2. **Creating a stand-alone interface** \- so far, the skill is good, but the conversation you were doing with the AI is always linear in the form of a chatbot. I had a discussion with a few of you, and one idea is that what if we have a separate ADHD interaction area? Something like ADHD skill + Miro/freeform. 3. **Large Scale Benchmarking** \- the next major thing that we are planning to do is run it on top benchmarks and see how the results are coming, because right now our sample size is comparatively low. Although we have come across so many individual reports of our project being tested on top evals, we need to update everything and put everything together. 4. **Finally, a community registrations for participate, talk, and contribute** \- I am opening up registrations for early adopters, contributors, members and maintainers of the project. If you want to join and build this thing, then you can just register via this link in comments. Would love your feedback peps. And congrats so so much for this journey ahead. Would talk to talk with you one on one.
The author introduces a novel approach for AI agents called ADHD, using parallel divergent ideation to enable non-linear thinking inspired by ADHD minds, though with higher cost and latency, and releases it as open source.
Claude Sonnet 5 is now generally available and rolling out in GitHub Copilot. Early testing shows strong coding performance, especially on CLI tasks, with good prompt-cache utilization and competitive latency.
A new open-source skill for coding agents that implements parallel divergent ideation—spawning isolated reasoning processes under distorted cognitive frames—to overcome premature convergence in autoregressive reasoning. The skill, available for Claude Code and other agents via npm, claims to significantly improve performance on creative and design tasks.