PaperConan Skill for Reviewing Research Source Data

Install in Claude Code

A local agent with shell access, Python 3.10 or later and read access to the data. The base CLI is enough for the synthetic CSV example. The all extra adds PDF and Word table extraction.

1. Install the skill using the repository's documented `npx skills add zixixr/paperconan -a claude-code -a codex` command. A pinned repository checkout can also supply the `skills/paperconan/` folder for a manual Claude Code installation.
2. Install the matching CLI with `python -m pip install paperconan==0.9.0` in a Python 3.10+ environment. Use `paperconan[all]==0.9.0` if PDF or Word extraction is needed.
3. Run `paperconan --version`, then reopen the agent session before requesting a scan.

Example task

Save this synthetic CSV as `fixture/replicates.csv`.

sample,run_a,run_b
s1,1.3,1.3
s2,3.7,3.7
s3,8.2,8.2
s4,14.6,14.6
s5,2.1,2.1
s6,9.4,9.4
s7,4.8,4.8
s8,6.5,6.5
s9,11.2,11.2
s10,18.7,18.7
s11,13.8,13.8
s12,20.3,20.3

Ask Claude Code to scan `fixture/` for duplicated numeric columns, or run `paperconan fixture/` and review `fixture/audit/scan.json`.

What to check

The default scan writes fixture/audit/scan.json and fixture/audit/report.html. In the source CSV, run_a and run_b are equal in every row. Check the file location, the actual detector output and scan coverage before describing a finding.

Original SKILL.md on GitHub · Updated 2026-11-14