Opening your workspace…
Start from past decisions in a file, or from the decisions your code already makes.
A spreadsheet (CSV or Excel), JSON or JSON Lines, one example per row with the right answer in one column. You check what it reads and decides, then train.
Drop the file here
Or try it with .
An LLM call that picks a label, a rule, a game or an agent choosing its next move. Your coding agent finds those decisions in your repository, logs real examples, measures Zils on them and trains a model where it falls short. It asks you before spending anything.
claude mcp add zils -- npx -y @zils/cli mcpnpx -y @zils/cli loginFind the decisions in this repo that Zils could make, measure Zils on them, and train a model for any it misses.