How do I install the Do.pdf To Markdown skill?

Run npx skillmds@latest add dull-bird/do-pdf-to-markdown in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Do.pdf To Markdown safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Do.pdf To Markdown?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Do.pdf To Markdown free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Do.pdf To Markdown?

dull-bird (@dull-bird) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Do.pdf To Markdown by dull-bird — Agent Skill

Name: Do.pdf To Markdown
Author: dull-bird

Do.pdf To Markdown

Convert PDF documents to high-fidelity Markdown with checkpointed page and section coverage verification. Use when the user asks to convert, transcribe, extract, or preserve a PDF as Markdown, especially for papers containing formulas, tables, figures, or multi-column layouts.

dull-bird f3051a4 2 files · 3.5 KB Updated Sep 17, 2026

File contents

PDF to Markdown

Preserve the source language and document structure. Use visual PDF reading for formulas, tables, and reading order; do not trust blind text extraction for complex layouts.

1. Build the manifest

Inspect the full PDF and write <name>_manifest.md beside it:

---
deeporbit_workflow: 1
workflow_id: pdf-<date>-<slug>
status: active
source: <name>.pdf
output: <name>.md
---

# Conversion Manifest
- Total pages: N
- [ ] Cover and metadata (p.1)
- [ ] Abstract (p.1-2)
- [ ] Section 1 (p.2-5)
- [ ] References (p.N-2-N)
- Tables: T
- Figures: F

Every page must belong to a checklist item. If bundled image extraction is available, run:

python <skill-dir>/scripts/extract_images.py <input.pdf>

PyMuPDF is optional and is used only for embedded images, never as the sole text source.

2. Convert with checkpoints

For each unchecked section:

Visually read its page range.
Append it to <name>.md in reading order.
Preserve headings, paragraph breaks, lists, fenced code, footnotes, tables, and emphasis.
Transcribe math as $...$ or $$...$$.
Reference extracted images with relative paths and source captions.
Compare paragraph, table, figure, and page coverage.
Mark that section checked only after its verification succeeds.

Continue in the same turn when feasible. If interrupted or context-limited, reread the manifest and resume at the first unchecked item. A runtime Goal or Tracker may mirror the overall objective, but the manifest remains authoritative. Do not use an external self-invoking loop.

3. Verify the whole document

All manifest entries are checked.
Every page is covered exactly once or its overlap is explained.
Output headings match the source section list.
Table and figure counts match.
Code, math, links, and footnotes are intact.
No source paragraphs were silently omitted.

Add a final conversion report with page, section, table, figure, image, and word counts. Keep the manifest when an audit trail is useful; otherwise ask before deleting it.

4. Final output

Write frontmatter at line 1:

---
type: note
title: "<Document Title>"
source: "[[<name>.pdf]]"
created: YYYY-MM-DD
tags: [pdf-conversion]
---

Set author: ai in frontmatter for every note you create; switch to author: mixed when substantially rewriting a human-authored note. Authorship lives in frontmatter only — never add visible badges. Save beside the PDF, link relevant existing Wiki concepts when appropriate, and use do.obsidian-open to present the result. Failure to open the desktop app does not invalidate the conversion.

---name: do-pdf-to-markdowndescription: Convert PDF documents to high-fidelity Markdown with checkpointed page and section coverage verification. Use when the user asks to convert, transcribe, extract, or preserve a PDF as Markdown, especially for papers containing formulas, tables, figures, or multi-column layouts.---# PDF to MarkdownPreserve the source language and document structure. Use visual PDF reading for formulas, tables, and reading order; do not trust blind text extraction for complex layouts.## 1. Build the manifestInspect the full PDF and write `<name>_manifest.md` beside it:```markdown---deeporbit_workflow: 1workflow_id: pdf-<date>-<slug>status: activesource: <name>.pdfoutput: <name>.md---# Conversion Manifest- Total pages: N- [ ] Cover and metadata (p.1)- [ ] Abstract (p.1-2)- [ ] Section 1 (p.2-5)- [ ] References (p.N-2-N)- Tables: T- Figures: F```Every page must belong to a checklist item. If bundled image extraction is available, run:```bashpython <skill-dir>/scripts/extract_images.py <input.pdf>```PyMuPDF is optional and is used only for embedded images, never as the sole text source.## 2. Convert with checkpointsFor each unchecked section:1. Visually read its page range.2. Append it to `<name>.md` in reading order.3. Preserve headings, paragraph breaks, lists, fenced code, footnotes, tables, and emphasis.4. Transcribe math as `$...$` or `$$...$$`.5. Reference extracted images with relative paths and source captions.6. Compare paragraph, table, figure, and page coverage.7. Mark that section checked only after its verification succeeds.Continue in the same turn when feasible. If interrupted or context-limited, reread the manifest and resume at the first unchecked item. A runtime Goal or Tracker may mirror the overall objective, but the manifest remains authoritative. Do not use an external self-invoking loop.## 3. Verify the whole document- All manifest entries are checked.- Every page is covered exactly once or its overlap is explained.- Output headings match the source section list.- Table and figure counts match.- Code, math, links, and footnotes are intact.- No source paragraphs were silently omitted.Add a final conversion report with page, section, table, figure, image, and word counts. Keep the manifest when an audit trail is useful; otherwise ask before deleting it.## 4. Final outputWrite frontmatter at line 1:```yaml---type: notetitle: "<Document Title>"source: "[[<name>.pdf]]"created: YYYY-MM-DDtags: [pdf-conversion]---```- Set `author: ai` in frontmatter for every note you create; switch to `author: mixed` when substantially rewriting a human-authored note. Authorship lives in frontmatter only — never add visible badges.Save beside the PDF, link relevant existing Wiki concepts when appropriate, and use `do.obsidian-open` to present the result. Failure to open the desktop app does not invalidate the conversion.

Frequently asked questions