Task-oriented tools that LLMs use to run complete ML workflows.
Each tool is one module under this package, declaring a ToolSpec alongside its executor. Everything the rest of the codebase consumes -- the schema dicts, the dispatch table, the MCP annotations, the notebook-export skip list -- is derived from the specs collected in SPECS below.
To add a tool: create the module with its ToolSpec, then import it here and append its spec to SPECS. That list is the registration point and is deliberately explicit -- it fixes tool ordering and makes a missing registration a visible omission rather than a silent one.
workflow/ train, predict, evaluate, benchmark, tune, plot, save_model data/ upload, profile, generate, read, preprocess, select_features analysis/ statistics discovery/ list, describe authoring/ agent-authored algorithms (skeleton, create, read, edit, ...) system/ install info, self-update, restart, REST serving notebook/ session export to .ipynb