What it does
The loop runs decide, tool, decide, tool until the goal is scored reached or it escalates, and a benchmark suite scores a run against hidden tests and a reference directory.
Agent tooling
jeffrey
A coding-agent CLI that splits the work: a decision model picks the next tool, scores progress and judges whether the goal is reached, while an LLM only fills in arguments and writes code.
Bookmark: jeffrey
Keep this in your collection.
Leave a noteWhat would you try with this?
: jeffrey
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
For a code change in your app, have a developer try this AI helper. The developer sets it up and limits how many steps it may take. One AI chooses what to do next; another writes the code. Ask the developer to check the changed files with you, even if the helper says the task is finished.