# Replication Checker

> Use when asked to verify reproducibility, audit a replication package, check a clean run, validate run instructions, or review before sharing research code.

- Skill: `meleantonio/replication-checker` (Agent Skill)
- Install (CLI): `npx skillmds@latest add meleantonio/replication-checker`
- Raw SKILL.md: https://api.skillmd.com/api/skills/meleantonio/replication-checker/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Security, Coding & Dev Tools
- Author: meleantonio (https://skillmd.com/u/meleantonio)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/meleantonio/replication-checker

---


# Replication Checker

This is a clean-room reproducibility checklist for economics research projects.

## Checklist

1. Read `AGENTS.md`, the example or project README, and any data source map.
2. Identify the documented entry point and the smallest safe verification command.
3. Check path hygiene: no `/Users/`, `C:\`, machine names, or hidden local data
   assumptions in runnable code.
4. Check dependencies: requirements are declared and match imports.
5. Check data documentation: source, unit of observation, sample restriction,
   transformations, and caveats are visible.
6. Check generated outputs: outputs are either reproducible or clearly ignored.
7. Run the smallest safe command only when the user or task allows execution.
8. Return GREEN only if runnable evidence supports it; otherwise use YELLOW or RED.

## Card-Krueger Baseline

Expected verification commands from the repository root:

```bash
python3 examples/card-krueger/src/did_analysis.py
python3 -m pytest examples/card-krueger/tests
```

## Output

```markdown
# Replication Report

**Status:** GREEN / YELLOW / RED

## Evidence

| Criterion | Status | Evidence |
| --- | --- | --- |

## Blockers

## Recommendations

## Commands
```

## Constraints

- Do not mark GREEN without command output or equivalent evidence.
- Do not modify data files during an audit.
- Do not treat a successful synthetic-data run as a substantive replication of
  Card and Krueger (1994).