Causal inference analysis

Skill SylphxAI/skills/skills/causal-inference-analysis

Define and evaluate the causal effect of an intervention, exposure, policy, or treatment on an outcome using explicit estimands, causal assumptions, study design, identification, diagnostics, and sensitivity analysis. Use for questions such as whether X causes Y, what would happen if X changed, observational or quasi-experimental impact analysis, confounding, mediation, selection, or causal attribution. Produce a Causal Inference Record. Do not use for ordinary correlation reporting, isolated software root-cause debugging, generic system maps, or merely configuring an A/B test.From its SKILL.md

Install

npx -y skills add SylphxAI/skills --skill causal-inference-analysis

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

24 days oldThe repository was created 24 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.8 KB, 632 tokens by cl100k_base, as published. Nobody here has run it

Causal Inference Analysis

Turn a causal claim into an explicit intervention, estimand, identification argument, and falsifiable analysis. Read references/causal-inference-methods.md before choosing a design or adjustment strategy.

Workflow

Define the intervention or exposure, comparator, population, outcome, time zero, follow-up horizon, and target estimand. Reject vague verbs such as “impact” until the counterfactual contrast is clear.
Establish temporal ordering and draw the causal assumptions. Distinguish confounders, mediators, colliders, selection mechanisms, measurement error, interference, and time-varying treatment or confounding.
Emulate the target experiment conceptually even when only observational data are available: eligibility, assignment strategy, follow-up, outcome, causal contrast, and analysis plan.
Select a design whose assumptions fit the data and assignment process. Prefer randomized evidence when ethical and feasible; otherwise justify adjustment, matching, target-trial emulation, difference-in-differences, regression discontinuity, instrumental variables, or another identified strategy rather than choosing by fashion.
State the identification assumptions and data requirements before analysis. Never adjust mechanically for every observed variable or condition on a mediator or collider without a causal reason.
Check overlap, balance, assignment and exposure integrity, missingness, model dependence, pre-trends or design-specific diagnostics, negative controls, alternative specifications, and sensitivity to unmeasured bias.
Report the estimate with uncertainty, assumptions, diagnostics, limits on transportability, and plausible alternative explanations. Not identified is a valid result when the evidence cannot support the causal claim.

Output

Produce a Causal Inference Record:

Causal question — intervention, comparator, population, outcome, horizon, and estimand.
Causal model — DAG or equivalent assumptions, timing, assignment, interference, measurement, and selection.
Identification and design — strategy, required assumptions, data, and why alternatives were rejected.
Analysis and diagnostics — estimate, uncertainty, robustness, negative controls, sensitivity, and unresolved threats.
Claim boundary — supported causal statement, non-supported statements, transportability, and evidence that would change the conclusion.

Boundaries