I built whatbroke, a CLI that diffs an AI agent's behavior between two runs of the same task.Swap a model or tweak a pro...

I built whatbroke, a CLI that diffs an AI agent's behavior between two runs of the same task.Swap a model or tweak a prompt, record a JSONL trace of each version, run `whatbroke diff`, and it reports dropped tool calls, drifted arguments, cost/latency moves, flipped outputs. Exit 1 on breaking changes, so it slots into CI.MIT, deterministic, fully offline — traces never leave your machine.https://github.com/arthi-arumugam-git/whatbroke#opensource #llm #cli

Read Original

Related