Small, sharp tools for AI agents and developer workflows that I’m building in public. Each one started from a real problem I hit, and I log the bugs, decisions, and tradeoffs as I go, not just the finished result.
Clinescope
Grades how well an AI coding agent actually did its job, by replaying a real run and scoring the moves it made. Most eval tools only check an AI's written answers, not whether it can build software.