Init Databricks Ml
Initialize or update a project with Databricks ML workflow scaffolding, skills, and agents. Use when starting a new ML project or adding missing Databricks integration to an existing project.
Handoff
Write or update a handoff document so the next agent with fresh context can continue this work.
Run On Databricks
Submit a training job to the Databricks GPU cluster, wait for results, and pull MLflow metrics back locally. Use when the user wants to train on GPU, run a full experiment, or execute code on the Databricks cluster. Triggers on "run on Databricks", "train on GPU", "submit job", "full training run".
Run Training On Databricks
Submit ML training to Databricks and pull results. Use when the user wants to run, train, or execute a model on Databricks.
Compare Runs
Compare MLflow experiment runs, rank them, and recommend next steps. Use when the user asks about results, which run is best, what improved, what hurt performance, or wants a summary of experiment history. Triggers on "compare runs", "what improved", "best model so far", "show me results", "which hyperparameters worked".
Research Papers
Research ML techniques, architectures, or papers relevant to the current task and summarize actionable insights. Use when the user asks "is there a better way to do X", "what are state-of-the-art methods for Y", "find papers on Z", "how do others solve this problem", or wants to understand recent advances in a technique.
Explore Data
Analyze a dataset. Use when the user wants to explore data, understand distributions, check quality, find correlations, profile a DataFrame, or generate EDA plots. Triggers on phrases like "look at the data", "what does the dataset look like", "check for missing values", "understand this dataset".
Check Results On Databricks
Analyze MLflow experiment results. Use when the user asks about model performance or experiment history.
Explain Code
Explains code with visual diagrams and analogies. Use when explaining how code works, teaching about a codebase, or when the user asks "how does this work?
Iterate
Full ML iteration cycle: implement a change, run on Databricks, compare results, suggest next step. Use when the user says "try X", "implement Y and test it", "iterate on the model", or "run an experiment with Z".
Train Local
Run a quick training experiment locally (CPU or MPS). Use when the user wants to test a model change, debug training code, run a smoke test, or validate a new architecture before submitting to Databricks. Fast iteration, small data, short epochs.