Opensourcing my EdgeResearch skill sets that I use about 30 times daily. You should to.
Heat trend
Collecting trend data
The percentage is based on available heat signal, not comment count or independent people.
This covers a coding tool or code-capability update — useful for developers assessing workflow changes and reusable value.
So guys, I've been dealing with a lot of agents shipping changes that I later found out made it worse for me and my project kept improving on something that didn't make no sense at all.
A few months ago I adopted Andrej Karpathy's RResearch repo and I used it quite a lot but it didn't really help me too much with development and planning. What I did is I've built a small set of skills on top of that repo and just turned it into an edge case research skill. This skill essentially does four things:
- It keeps it or throws it away, very similar to Karpathy's other research but it's more for prompts, copy, code, and system architecture.
Here is what made it work for me. This skill actively refuses most common agent lies, like "I think this is better," "kills and refuses stuff like just rewrite and rust," or "tests are green without a proof" or "this branch passes locally without a proof."
It's MIT-licensed, works as a one-shot install in Claude Code / Codex / Hermes / OpenClaw. I added a few worked examples showing the "I just say 'EdgeResearch {something}' and the agent does the rest" UX.
This is my first public repo, so be gentle, would love any meaningful contributions, too.
Go try it, it takes a few minutes usually (depening on a model and a task), and you'll see how deep and detailed the plans get, and how much more useful edge case coverage you'll get.