analytic-question-framing - SKILL.md Agent Skill

name: analytic-question-framing description: Refine vague data science requests into precise, answerable analytic questions. Use when a user asks what analysis to run, provides a broad business or research goal, needs to classify a question as descriptive, exploratory, inferential, predictive, causal, or mechanistic, or needs acceptance criteria before data cleaning, EDA, modeling, or reporting.

Analytic Question Framing

Start every analysis by making the question answerable with the available data.

Domain Context Requirement

If a Domain Context Contract is available, use it as controlling context. Align the question with the stated domain, stakeholder decision, unit of analysis, target/KPI, success metric, constraints, and prohibited claims. If no contract exists for a business/domain request, invoke or request domain-problem-interviewer-researcher before finalizing the question.

Question Types

Descriptive: summarize what is in a dataset.
Exploratory: search for patterns, hypotheses, relationships, or anomalies.
Inferential: estimate population relationships from a sample.
Predictive: predict an outcome for new units.
Causal: estimate what changes when an intervention changes.
Mechanistic: explain the process that creates an observed effect.

Procedure

Restate the relevant Domain Context Contract fields or create a provisional contract if missing.
Rewrite the request as one primary question and at most three follow-up questions.
Name the question type. If mixed, split it into separate questions.
Identify the unit of analysis, population, time window, outcome, predictors, grouping variables, and decision threshold.
Check whether the available data can answer the question. If not, state the nearest answerable question.
Define success criteria: required table, plot, model, metric, confidence/uncertainty statement, or decision.
State disallowed interpretations, especially causal claims from non-causal data.

Output Template

Domain context used:
Primary question:
Question type:
Unit of analysis:
Population/scope:
Outcome:
Inputs/features:
Time window:
Minimum viable analysis:
Risks and non-claims:
Acceptance criteria: