Snapshot Data
Generate a versionable YAML manifest next to a dataset capturing SHA-256 hash, schema fingerprint, shape, and column metadata. Used to pin a dataset version so future re-runs can verify they are working on the same data.
Freeze Deps
Detect imports in a Jupyter notebook, resolve their installed versions, and emit either requirements.txt (pip) or a poetry add command (Poetry projects). Also appends a watermark cell to the notebook for inline version auditability.
Quality Report
Generate a one-page data quality triage report with traffic-light scores for completeness, consistency, and usability. Outputs actionable blockers and cleaning recommendations before full EDA.
Eda Report
Synthesize an executed and annotated EDA notebook into a 1-3 page executive markdown report. Selects 3-6 key findings, removes code, structures into 7 canonical sections, and produces a stakeholder-ready document. Closes the EDA pipeline after /annotate-findings.
Annotate Findings
Read an executed Jupyter notebook and rewrite placeholder observation cells (marked `<!-- annotate-findings: pending -->`) with real findings derived from the actual cell outputs. Closes the loop after /execute-notebook so observations reference real numbers, not pre-execution speculation.
Analyze Dataset
Generate a structured analytical Jupyter notebook from any dataset path. Produces EDA, statistical tests, visualizations, and optional modeling following notebook best practices.
Execute Notebook
Execute a Jupyter notebook headless with kernel restart using papermill. Captures all cell outputs and overwrites the notebook with the executed version. Fails loudly on the first error with cell index and traceback.