zpower426

Dispatch as a subagent to execute a specific analysis task. Receives full task specification from the orchestrating agent. Does NOT inherit session context. Examples: <example>Context: An orchestrating agent is running subagent-driven-analysis. user: "Execute Task 3: numeric feature preprocessing" assistant: "Dispatching analyst subagent with the full task specification." <commentary>Each task gets a fresh analyst subagent with only the context it needs.</commentary></example>

Use after statistical review has passed to check code quality of analysis code. Reviews for vectorization, reproducibility, clarity, and artifact correctness. Examples: <example>Context: Statistical review has passed for a feature engineering task. user: "Statistical review approved the feature engineering code" assistant: "Now let me dispatch the code quality reviewer to check implementation quality." <commentary>Code quality review always comes after statistical review, never before.</commentary></example>

Use after an analyst subagent completes a task to verify statistical correctness. Reviews for data leakage, correct metric selection, proper cross-validation, and sound statistical methodology. Examples: <example>Context: A feature engineering task has been completed. user: "Feature engineering for numeric columns is done" assistant: "Let me dispatch the statistical reviewer to check for leakage and correct transformer usage." <commentary>Statistical review must happen before code quality review, and always after any data transformation or modeling task.</commentary></example>

Start a new analysis project with structured hypothesis-first brainstorming

Execute an analysis plan using subagent-driven analysis with two-stage review

Break an approved analysis design into discrete, executable tasks

Use at the START of every analysis session and at the END of every skill stage. Maintains artifacts/analysis_manifest.json as the single source of truth for analysis state, preventing context drift across long sessions.

Use BEFORE any analysis project, ML task, or significant data work. Explores business context, data characteristics, and analytical approach before any code or modeling.

Use when exploring any dataset for the first time, or when asked to perform EDA. Enforces systematic exploration before any modeling or transformation.

Use BEFORE dispatching any subagent that needs to understand the dataset. Generates a high-density, PII-free data profile in Markdown so subagents receive structured context instead of raw data.

Use before any model training, feature engineering, or data transformation. Validates data schema, quality constraints, and statistical expectations.

Use when a data pipeline, model, or analysis produces unexpected results, errors, or performance degradation. Enforces root cause investigation before any fix.

Use when executing a written analysis plan task by task. Manages task state, enforces two-stage review (statistical first, then code quality), and gates manifest updates behind completed reviews.

Use when creating, transforming, or selecting features. Enforces leakage-free, reproducible feature pipelines.

Use when analysis is complete, all validations pass, and you need to decide how to integrate the work — guides delivery of analysis artifacts via commit, PR, or archive.

Use whenever building features for time-series or any temporal dataset. Enforces strict temporal integrity: no future data in features, no post-event information, correct CV strategy.

Use when evaluating a trained model. Enforces one-time test set evaluation, statistical significance testing, and calibration checks.

Use when choosing which model(s) to train. Enforces baseline comparison before hyperparameter tuning, and correct metric selection for the task type.

Use when writing analysis reports, stakeholder summaries, or presenting model results. Enforces reproducibility, honest uncertainty communication, and actionable conclusions.

Use after completing any major analytical task (EDA, features, modeling). Enforces rigorous auditing for target leakage, metric choice, and statistical significance.

Use when executing an analysis plan with independent tasks. Dispatches fresh subagents per task with two-stage review: statistical correctness first, then code quality.

Use before any training step to enforce three-layer data validation: Physical (schema), Logical (business rules), Statistical (distribution drift). Blocks training if any CRITICAL assertion fails.

Start here. Introduction to the skills library for data mining and statistical rigor.

Use when any analysis, model, or report is claimed to be complete. Runs mandatory artifact integrity, statistical evidence, and reproducibility checks before any delivery.

Use after brainstorming to break an approved analytical design into executable tasks. Each task must be completable in 15-30 minutes with full specificity.

Use when creating a new datapowers skill. Defines the required structure, Statistical Pressure Test protocol, Iron Law format, and adversarial testing methodology.