Skip to main content
skill CX Codex

pdf-to-markdown · claesbackman — Claude Code skill: what it does, how to install

pdf-to-markdown splits PDF documents into page blocks and converts them to Markdown, automatically trimming bibliographies.

MIT ★ 579 updated 2026-10-28

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add claesbackman/AI-research-feedback --skill pdf-to-markdown -a claude-code
Open on GitHub

What pdf-to-markdown does

This skill turns PDF files into structured Markdown alongside the original document. It prioritizes the pdftotext utility from the poppler library, running an awk filter directly against layout text. That pipeline generates 20-page bracket markers and clips trailing sections when encountering headers like "References", "Bibliography", or "Appendix".

When pdftotext is missing, the skill falls back to the native Read tool in 20-page segments. Without poppler installed, large PDFs process much slower and risk hitting subagent idle timeouts during long extraction passes.

What it helps with

  • convert academic PDF papers to readable Markdown
  • strip reference sections and appendices from documents
  • paginate long PDF extracts with clean block markers
  • extract research text without consuming agent context windows

Example prompts for the agent

  • Convert paper.pdf to markdown text
  • Run pdf-to-markdown on docs/report.pdf
  • Extract the main text from manuscript.pdf into markdown

How to install pdf-to-markdown

1Skills CLI (Vercel)

npx skills add claesbackman/AI-research-feedback --skill pdf-to-markdown -a claude-code

2Manually: put the files in place

~/.claude/skills/pdf-to-markdown/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/pdf-to-markdown/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about pdf-to-markdown

How to install pdf-to-markdown in Claude Code?

Run npx skills add claesbackman/AI-research-feedback --skill pdf-to-markdown -a claude-code in your terminal.

What does pdf-to-markdown do?

It converts PDF files into Markdown text files while adding page interval markers and dropping terminal reference sections.

Does pdf-to-markdown require external tools?

It works best with pdftotext installed on your machine; without it, the workflow falls back to chunked reading via the internal Read tool.

Source text

SKILL.md · MIT
---
name: pdf-to-markdown
description: Split a PDF into chunks and convert it to readable markdown text. Use when the user wants to read, extract, or convert a PDF document.
user-invocable: true
argument-hint: [path/to/file.pdf]
allowed-tools: Read, Bash(mdls *), Bash(pdftotext *), Bash(which *), Write
---

# PDF Split & Convert

Convert a PDF file to readable markdown text. Handles large PDFs efficiently.

## Input

- `$ARGUMENTS[0]` — Path to the PDF file (required)

## Choose the method first

Before reading anything, check whether `pdftotext` (part of poppler) is available:

```bash
which pdftotext
```

- **If available** → use the **pdftotext path** below. It is 10–100× faster than the Read tool, uses no model context for the text content, and doesn't suffer from stream idle timeouts. **Always prefer this for PDFs longer than ~30 pages.**
- **If not available** → fall back to the **Read-tool path**. Warn the user that large PDFs (>40 pages) may hit stream idle timeouts when run in subagents (~12 min cap). Prefer running in the main conversation for large files.

## Resolve the path and get page count

1. If the path is relative, resolve it relative to the current working directory.
2. Run `mdls -name kMDItemNumberOfPages "<pdf_path>"` to get the total page count. If `mdls` is unavailable or returns `(null)`, use `pdftotext` or Read to probe.

## Method A — pdftotext (preferred)

Extract the full PDF to text in one shot, then trim at references/appendix and add page markers. `pdftotext` emits a form-feed character (`\f`) at every page break — use that for pagination.

Reference implementation (bash + awk):

```bash
PDF="$1"
OUT="${PDF%.pdf}.md"
TITLE=$(basename "${PDF%.pdf}")
TMPTXT=$(mktemp)

pdftotext -layout "$PDF" "$TMPTXT"

awk -v title="$TITLE" '
BEGIN {
    print "# " title
    print ""
    page = 1
    printf "---\n## Pages %d-%d\n---\n", page, page+19
    next_marker = page + 20
}
{
    # Convert form-feed page breaks to newlines and count pages
    n = gsub(/\f/, "\n")
    if (n > 0) {
        page += n
        if (page >= next_marker) {
            printf "\n---\n## Pages %d-%d\n---\n", next_marker, next_marker+19
            next_marker += 20
        }
    }

    # Build a stripped copy for heading detection.
    # CRITICAL: strip both form feeds AND embedded newlines — gsub above inserts
    # newlines into $0, which will defeat regex anchors like ^ and $ if you skip this.
    stripped = $0
    gsub(/[\f\n]/, "", stripped)
    sub(/^[ \t]+/, "", stripped)
    sub(/[ \t]+$/, "", stripped)

    if (length(stripped) > 0) {
        # References / Bibliography — standalone word, optionally numbered, short, no prose punctuation
        if (length(stripped) < 50 && stripped !~ /[(),;]/) {
            if (stripped ~ /^([0-9]+\.?[ \t]+)?(References|REFERENCES|Bibliography|BIBLIOGRAPHY)$/) exit
            if (stripped == "Works Cited") exit
        }

        # Appendix — length up to ~120 chars (some titles are long), no prose punctuation
        if (length(stripped) < 120 && stripped !~ /[(),;]/) {
            # "Appendix A" alone (bare letter, no title)
            if (stripped ~ /^Appendix[ \t]+[A-Z][0-9]*$/) exit
            # "Appendix A. Title" or "Appendix A: Title" — punctuation REQUIRED to avoid
            # matching body-text references like "Appendix H examines the effect..."
            if (stripped ~ /^Appendix[ \t]+[A-Z][0-9]*[.:][ \t]+[A-Z].*$/) exit
            # "APPENDIX A" variants
            if (stripped ~ /^APPENDIX[ \t]+[A-Z][0-9]*([ \t]+.*)?$/) exit
            # "Online Appendix [A]"
            if (stripped ~ /^Online[ \t]+Appendix([ \t]+[A-Z].*)?$/) exit
            # "Supplemental/Supplementary/Internet Appendix"
            if (stripped ~ /^(Supplement(al|ary)|Internet)[ \t]+Appendix([ \t]+.*)?$/) exit
        }
    }

    print
}
' "$TMPTXT" > "$OUT"

rm -f "$TMPTXT"
echo "Wrote $OUT ($(wc -l < "$OUT") lines)"
```

After running, **sanity-check the output**:

```bash
# Should print nothing if trimming worked
grep -cE "^[[:space:]]*(References|Bibliography|REFERENCES|BIBLIOGRAPHY)[[:space:]]*$" "$OUT"

# Eyeball last few lines — should be prose/conclusion, not body-of-table or mid-paragraph
tail -5 "$OUT"
```

If the tail looks truncated mid-paragraph, the heading detection likely fired on a false positive. If the tail shows references or appendix content, the detection missed the heading — inspect the PDF text around that area and extend the regex.

## Method B — Read tool (fallback when pdftotext unavailable)

1. Read the PDF in chunks of up to 20 pages at a time using the Read tool's `pages` parameter: 1-20, 21-40, 41-60, etc.
2. Focus on the **MAIN TEXT ONLY**. Stop including content once you hit "References", "Bibliography", "Works Cited", or an appendix section. If references appear mid-chunk, keep everything before them and drop the rest.
3. Compile output:
   - Save alongside the PDF with a `.md` extension.
   - Header: `# [Original Filename]`
   - Page markers between chunks: `---\n## Pages X-Y\n---`
   - Preserve extracted text as-is.

**Warning:** The Read tool is slow for large PDFs (roughly 30–60 seconds per 20-page chunk). A 60-page paper can take 3–4 minutes, and sub-agents have a ~13-minute stream idle timeout that this can hit. When running a batch of conversions, run them sequentially in the main conversation or use Method A.

## Heading-detection pitfalls (hard-won lessons)

These false positives broke earlier attempts — keep them in mind whether you use Method A or B:

- **Parenthetical references in body text** like `(see Appendix B.5)` — exclude lines containing `(`, `)`, `,`, or `;`.
- **Body text starting with "Appendix X ..."** like `Appendix H examines the differential effect...` — require punctuation (`.` or `:`) immediately after the appendix letter when a title follows. Bare "Appendix A" alone on a line is still valid.
- **Line-wrapped headings** — `pdftotext` can wrap `Online Appendix` across two lines if the PDF's layout is unusual. You'll see `Online` on one line and `Appendix` on the next. The regex above matches the joined form; if you see false trims at a lone `Appendix` line, inspect and tighten.
- **Form-feed at start of page** — `pdftotext` emits `\f` as the first character on every new page. After `gsub(/\f/, "\n")` on `$0`, the line has an embedded `\n` that defeats `^` / `$` anchors unless you also `gsub(/[\f\n]/, "", stripped)` on your detection copy.
- **Length thresholds** — simple "< 50 chars" is too tight for appendix titles. "Appendix A. Merging Mortgages with the Real Estate Database" is 59 chars. Use ~120 for appendix patterns, ~50 for bare References.
- **Numbered section headings** — some papers format as `7 References` or `7. References`. Allow an optional leading number.

## Report results

Tell the user:
- Method used (pdftotext vs Read tool)
- Total pages processed
- Output file path and final line count
- Where trimming occurred (last section / page number included)
- Any pages that were unreadable or empty

For batch conversions, print a summary table and note any files whose trim point looks suspicious (very short output, or output ending mid-sentence).

© claesbackman · MIT · claesbackman/AI-research-feedback

Similar skills

All skills for Writing and docs →

pdf-to-markdown by other authors comes first.

pdf-to-markdown PSPDFKit-labs pdf-to-markdown turns a PDF into structured Markdown with headings, tables, lists and reading order intact. Handy when an agent needs the text of a document to work with. CX Codex +1 870installs pdf-to-markdown aliceisjustplaying pdf-to-markdown loads an entire PDF into the agent's context as structured Markdown with tables and images instead of grep-sized fragments. CX Codex +1 14installs pdf-to-markdown lttr pdf-to-markdown transcribes scanned and image-based PDFs into formatted Markdown files saved in the same directory as the source document. CX Codex +1 5installs PDF to Markdown dudko-dev The PDF to Markdown MCP server connects an AI agent to the pdf2md API and prepares requests to turn PDFs into Markdown. Sign in with a dudko.dev account to use it. CX Codex +1 pdf official openai A skill for reading, creating and reviewing PDFs where layout matters. Useful when the agent builds a document and needs to see how it renders. CX Codex +1 14,000installs photo-to-scanned-pdf daymade photo-to-scanned-pdf converts phone photos of paper pages or digital drafts into scanner-look A4 PDFs with perspective correction and clean backgrounds. CX Codex +1 358installs openai-pdf OpenAI openai-pdf guides Claude Code in reading, building, and visually validating PDF documents. ★ 540repo stars pdf-reading SeifBenayed pdf-reading — a skill for reading PDF files from disk: text extraction, page rasterization, table parsing, image and attachment retrieval. CX Codex +1 ★ 116repo stars convert-documents-to-markdown firecrawl convert-documents-to-markdown converts office documents, spreadsheets, and PDFs into GitHub-Flavored Markdown via anydoc. Useful when an agent must read DOCX or XLSX files without manual copy-pasting. CX Codex +1 13,700installs

From the same set13

The whole set →
audit-analysis audit-analysis launches an isolated subagent that adversarially checks changed empirical analysis code against a base branch, catching errors before a referee does. CX Codex +1 ★ 579repo stars explain-diff The explain-diff skill turns a code-change review into a standalone HTML page with a quiz. Use it to explain a diff and examine its effects on empirical results. CX Codex +1 ★ 579repo stars review-paper-light review-paper-light is a Claude Code skill that checks economics papers before submission, assessing contribution, identification, and causal claims in about a minute. CX Codex +1 ★ 579repo stars review-paper-code review-paper-code reviews research papers in LaTeX alongside Stata, R, or Python code, verifying reproducibility and checking empirical claims against scripts. CX Codex +1 ★ 579repo stars explorable-deck The explorable-deck skill turns a paper, notes, or a codebase into a Quarto reveal.js talk with a clear narrative and live SVG stages. Use it for lectures, conference talks, or presenting research. CX Codex +1 ★ 579repo stars explorable-beamer explorable-beamer builds LaTeX Beamer decks in Nicky Case's explorable-explanation style: one idea per frame, assertion titles, a running example, TikZ stages the presenter walks through, and an appendix of side rooms. CX Codex +1 ★ 579repo stars explorable-pptx explorable-pptx generates explorable-style PowerPoint slide decks with per-frame speaker notes and automated screenshot validation. CX Codex +1 ★ 579repo stars paper-version paper-version turns a LaTeX research paper into a public-facing brief or summary and produces an HTML page. Use it to explain a study without assuming readers know economics. CX Codex +1 ★ 579repo stars

Project sponsors

Companies that keep this analytics open