topic-feasibility-screener is a Codex skill for screening empirical research topics from existing .csv or Stata .dta datasets. It combines two checks:

1. Theory and literature feasibility: whether the topic has a clear research question, literature gap, plausible mechanism and defensible boundary.
2. Data feasibility: whether the dataset contains usable outcomes, key variables, controls, variation and enough observations for a minimal empirical test.

The skill is designed for early-stage thesis, accounting, finance, economics and management paper development. It helps prioritize topics before committing to a full empirical design.

• Reads .csv and .dta files.
• Profiles variables, missingness, distributions and likely roles.
• Detects candidate outcomes, treatment/key variables, controls, identifiers and time variables.
• Runs minimal OLS-style screening models with robust standard errors.
• Ranks candidate topic directions by statistical signal and feasibility.
• Probes literature density and recency through an OpenAlex fallback script.
• Produces Markdown and CSV reports for review.

Copy this folder into your Codex skills directory:

C:\Users\<you>\.codex\skills\topic-feasibility-screener

Then invoke it in Codex:

[$topic-feasibility-screener] Here is my DTA/CSV. The rough topic is digital transformation and financing constraints. Please screen feasible topics and run minimal validation.

Run the data screener directly:

python scripts\screen_topics.py "path\to\data.dta" --topic "data-resource recognition and corporate reputation" --json

With specified variables:

python scripts\screen_topics.py "path\to\data.dta" `
  --topic "data-resource recognition and corporate reputation" `
  --y ReputationScore `
  --x entry `
  --controls Size Lev ROA Growth BM Board Indep Top1 SOE Cashflow Loss `
  --id firm_id `
  --time year `
  --json

Run the literature fallback probe:

python scripts\literature_probe.py `
  --topic "data resource recognition corporate reputation" `
  --terms "data assets" "accounting recognition" "corporate reputation" `
  --outdir ".\literature_probe"

The data screener creates:

• variable_profile.csv
• candidate_models.csv
• topic_screening_report.md
• summary.json when --json is used

The literature probe creates:

• literature_probe.csv
• literature_probe_report.md
• literature_probe_summary.json

This skill is for topic screening, not final causal inference.

• A significant coefficient is not proof of causality.
• A topic is not feasible unless both theory and data are plausible.
• DID, IV, mechanism and DML designs still require separate identification checks.
• Literature-search failures should not be interpreted as absence of literature. Use CrossRef, Semantic Scholar, Google Scholar or a dedicated academic-search workflow for final verification.

MIT License.