Hi!
I sat down and created a repo: https://github.com/planetis-m/skills, it contains SKILL files for agents writing Nim code. Currently, the following are included:
For me, nim-c-bindings, nim-c-wrappers, nim-ownership-hooks, and nim-style-guide have been invaluable. The rest still need more work.
These SKILLs focus on things agents often get wrong—like writing correct destructors—so give them a try and personalize them to your needs.
How to contribute: Fork the repo, clone it to ~/.agents/skills, and make your own additions or improvements. It’s public domain, no CoC, no restrictions.
If you want to help further, the best approach is: during your session, when the agent messes something up, instruct it to collect an example of what went wrong and how it should look instead, then submit that as a bug report.
Cheers!
Since vacations have started, I’m keeping my coding plan active by making the development of skill files more systematic: https://github.com/planetis-m/skills_experiment. An openclaw agent is running with GLM 5.1 to review every claim made in the original skill files and ground them with tests as evidence. The verified new skill is then benchmarked against the original and rated blindly by a judge sub-agent, which is quickly exhausting the rate limits (in a productive way). I also have Codex review the progress in order to avoid concept drift and inefficiencies.
What I’m particularly interested in is answering the question: what is idiomatic Nim code? If you can point me to code samples or projects that you consider good examples, I would appreciate it. I’ve already fed most of my own code into the system to identify patterns, but additional sources would help diversify the dataset.