W W-Publication-Auditor

Publication-safety audit for Codex workflows

Trace every claim from data to final publication.

W-Publication-Auditor reviews manuscripts, response letters, SI, Source Data, figures, and analysis code for alignment gaps before submission.

Audit flow from data and code to manuscript and response letter
Text alignment response letter, manuscript, SI, captions
Data lineage Source Data, CSV, code outputs, figure claims
Word review red problem text and anchored comments

Workflow

Designed for the revision-to-publication gap.

01

Inventory artifacts

Find manuscripts, SI, response letters, source workbooks, generated outputs, figures, and code.

02

Extract evidence

Parse Word paragraphs, table rows, highlighted text, captions, workbook sheets, and source-data ranges.

03

Align claims

Compare response promises with revised text and classify missing, stale, near-match, and overclaim cases.

04

Annotate review copy

Create Word copies with red problem text and comments that include the exact error line, evidence, and fix.

Local environment

Install with uv, then run auditable commands.

The skill keeps the agent focused on publication-safety logic. The Python tools handle deterministic file parsing, workbook inspection, sentence matching, and Word comment insertion.

bash skill/w-publication-auditor/scripts/setup_uv.sh
source .venv/bin/activate
wpa-inventory ./paper-package --out audit_outputs
wpa-extract-docx response.docx --out audit_outputs/letter_blocks.csv
wpa-align-text audit_outputs/letter_blocks.csv audit_outputs/manuscript_blocks.csv \
  --out audit_outputs/sentence_alignment.csv

Skill package

Use it with open-source Codex + model + skills.

Download the skill zip, unpack it into your Codex skills directory, and let the agent call the bundled scripts during audits.

Download w-publication-auditor-skill.zip