# Paper Reproduction

> Reproduce ML/AI papers by turning claims into runnable notebooks, metrics, and traceable gaps. Use for paper replication plans and reproduction reports.

- Skill: `lanma-ctrl/paper-reproduction` (Agent Skill)
- Install (CLI): `npx skillmds@latest add lanma-ctrl/paper-reproduction`
- Raw SKILL.md: https://api.skillmd.com/api/skills/lanma-ctrl/paper-reproduction/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: Lanma-ctrl (https://skillmd.com/u/lanma-ctrl)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/lanma-ctrl/paper-reproduction

---


# Paper Reproduction Skill

Use this skill when reproducing a paper or blog-post experiment.

## Workflow
1. Capture paper metadata: title, link, claimed contribution, target metric.
2. Convert the paper into a small reproduction checklist:
   - dataset
   - baseline
   - implementation assumptions
   - evaluation metric
   - expected result
3. Start with the smallest runnable notebook before scaling up.
4. Record deviations from the original paper explicitly.
5. Produce a short reproduction report: what matched, what did not, and likely reasons.

## For this demo
Use a toy synthetic regression task instead of a real paper so the full loop is quick and local.