| name | replication-analysis |
| description | Implement build and analysis work inside this replication template. Use when Codex needs to create or revise scripts under `code/build/` or `code/analysis/`, construct datasets, run estimators, generate tables or figures, or keep outputs aligned with the numbered issue-folder workflow. |
Replication Analysis
Use this skill to turn an approved analysis idea into scripts and outputs that fit the repo's numbered workflow.
Workflow
- Locate the target workstream and nearest existing pattern.
- Decide whether the task belongs in:
code/build/NN_issue_name/
code/analysis/NN_issue_name/
- Implement the smallest vertical slice that produces a concrete artifact.
- Save outputs to the matching numbered folder under
data/build/ or output/.
- Verify with the narrowest credible command, preferably through
nix-shell.
Non-negotiables
- Use relative paths only.
- For R execution, prefer
nix-shell --run "Rscript ..." unless the task is explicitly about bootstrapping the environment itself.
- For Python execution, prefer
uv run ... and keep dependency changes reflected in the repo's Python dependency files.
- Document sample restrictions, merges, and dropped observations.
- Save reproducible intermediates when they matter for downstream work.
- Do not silently change estimators, clustering, or file destinations.
Load references as needed
- Read
references/script-standards.md for script structure, output conventions, and verification expectations.