| name | clean-research-report |
| description | Strips citation artifacts and hidden Unicode characters from ChatGPT deep research exports (.md or .docx). Use when the user has a .md or .docx file from ChatGPT deep research that contains leftover citation tokens (citeturnX, fileciteturnX), entity[] tags, garbled Unicode characters, or a trailing references dump. Also triggers when user mentions "clean report", "remove citations", "fix deep research export", or provides a .docx deep research file. |
clean-research-report
Cleans ChatGPT deep research exports. Accepts both .md and .docx files.
For .docx input, the script converts to GitHub-flavored markdown via pandoc first (must be installed), then applies the full cleaning pipeline. The .docx format is preferred because ChatGPT's markdown export strips reference hyperlinks while the Word export preserves them as inline links.
What it removes
- Citation tokens:
citeturn0search0, fileciteturn14file18, chained variants
- Private-use Unicode delimiters:
\ue200, \ue201, \ue202 (hidden in text)
- Entity annotation tags:
entity["company","Google","internet company"]
- Trailing references/bibliography section (the URL dump after the last horizontal rule)
- Resulting double spaces and trailing whitespace
What it preserves
- Inline citation hyperlinks like
[\[1\]](URL) in the body text (these are the actual useful source links)
Usage
uv run {baseDir}/scripts/clean_report.py report.docx
uv run {baseDir}/scripts/clean_report.py input.md
uv run {baseDir}/scripts/clean_report.py input.docx output.md
uv run {baseDir}/scripts/clean_report.py input.md --inplace
Prerequisites
Workflow
- Identify the file the user wants cleaned (
.md or .docx)
- Run:
uv run {baseDir}/scripts/clean_report.py <input> [output | --inplace]
- Report where the cleaned
.md file was written
