Convert a ChatGPT Report Export to Markdown with Codex

Compare Markdown, DOCX, or PDF copies of one report. Recover verifiable links and footnotes, clean up export artifacts, and flag time-sensitive claims for review.

中文

Install in Codex

Open the pinned project in Codex and keep its normalization reference available. Use local document tools that can read the source formats and hyperlink relationships.

Run git clone https://github.com/oaknational/oak-python-starter.git. From the checkout, run git checkout --detach 11113a35c0c5be3f9a3b55397693c6de248e5ed4.
Keep the original checkout and .agent/memory/active/patterns/chatgpt-report-normalisation.md. Copy the entire .agent/skills/chatgpt-report-normalisation folder into .agents/skills/chatgpt-report-normalisation in the checkout.
Open the checkout in Codex. Select the skill via /skills or $chatgpt-report-normalisation, then provide copies you can use and identify the headings and citations to preserve.

Example task

Make two copies of a fictional report. Put an export citation marker and two heading levels in Markdown. In the DOCX, make the words “example source” in that sentence link to https://example.com/. Keep the heading levels and restore the citation as a footnote.

What to check

The result keeps the headings, defines the body footnote, and points to the recovered URL. Unresolved citations are listed with their locations.

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

Frequently asked question

Can I recover all source URLs from the Markdown alone?

Only where you can establish the URL. If Markdown lost a link, check DOCX hyperlink relationships or another original copy. Record anything that still cannot be resolved.