I build small tools that help AI coding agents work together, and I send fixes to the open-source tools I use every day.
New: look-before-you-code, a skill and a pre-write hook that make Claude Code and Codex search for existing solutions before writing new code.
New: prove-it-review, a skill and a reviewer subagent that make coding agents prove a delivery with fresh evidence before calling it done.
New: subagent-tiers, reusable Claude Code agent definitions with a guide for matching work to evidence, risk and review needs.
Now: a message box that lets Claude Code and Codex sessions hand work to each other and wake each other up. Open-sourcing it soon.
Recently merged
- Automattic/harper#4991: skip conventional commit prefixes when linting commit messages
- vyncint/termlens#793: detect JSON screens after leading whitespace
- pymeasure/pymeasure#1595: clarify how to clone your fork
I'm based in Turkey and also work on Turkish language tools at KarıncaAI.
I build with Claude Code and Codex: I set the rules and make the calls, the agents write the code, and every change is tested before it goes out. Each project says exactly who did what in its AI-USE file.
