What problem does it solve? Running ML experiments manually requires writing training code, executing it safely, and tracking metrics by hand. This Skill automates the full experiment loop inside an isolated sandbox so every trial is reproducible and logged. ## Core Features & Use Cases - Dynamic Script Generation: Writes training and evaluation code into workspace/experiment_{trial_id}.py based on a hypothesis specification and dataset target. - Sandboxed Execution: Runs each experiment inside TrueForge's isolated container sandbox within a 2-minute time budget. - Structured Metrics Logging: Parses val_loss, val_acc, val_f1, and val_auc and appends each trial as a row to workspace/results.tsv, marking regressed trials as DISCARDED. - Use Case: A research agent proposes a new model configuration; the Skill generates the training script, runs it in the sandbox, and appends the resulting metrics to results.tsv for downstream plotting and paper writing. ## Quick Start Run an experiment for the given hypothesis and dataset target, then append the validation metrics to workspace/results.tsv.
