processing-pdfs
The processing-pdfs skill handles PDF work in your workspace. It extracts text and tables, merges and splits files, creates new PDFs, fills forms, adds watermarks or annotations, and reads scanned pages with OCR. It also processes large files and whole folders of PDFs in a way that keeps Claude’s context window from filling up.
When Claude Uses It
Section titled “When Claude Uses It”Claude loads the skill when you ask it to process a PDF, extract data from one, merge or split pages, or fill in a PDF form. A request such as “Review the PDFs in this folder for competitive analysis” starts the bulk workflow.
The Size Check Comes First
Section titled “The Size Check Comes First”Before it reads any PDF, Claude checks the file’s size and page count and places it on the five-level PDF Processing Scale. This step is mandatory, because a large file read in one pass fails and wastes a turn.
| Scale | Size | Pages | Approach |
|---|---|---|---|
| 1 Trivial | Under 1MB | Under 10 | Read directly |
| 2 Small | 1 to 5MB | 10 to 50 | Read directly and write a summary to disk |
| 3 Medium | 5 to 20MB | 50 to 200 | Read 20 pages at a time, writing findings after each chunk |
| 4 Large | 20 to 50MB | 200 to 500 | Split the file with qpdf first, then read in chunks |
| 5 Massive | Over 50MB | Over 500 | Split with qpdf and hand the chunks to parallel agents, each writing to disk |
When size and page count point to different levels, Claude uses the higher one. A 3MB file with 180 pages is Scale 3. At Scale 3 and above, Claude reads the first five pages to find the structure, then reads only the sections it needs. The scale itself is a rule. See PDF Processing Scale.
Tools It Uses
Section titled “Tools It Uses”| Task | Tool |
|---|---|
| Extract text | pdfplumber |
| Extract tables | pdfplumber |
| Merge or split | pypdf |
| Create a PDF | reportlab |
| Read scanned pages | pytesseract, after converting pages to images |
| Split large files | the qpdf command-line tool |
Processing a Folder of PDFs
Section titled “Processing a Folder of PDFs”For three or more PDFs, the skill uses a method it calls Extract-Summarise-Release.
- Claude asks what you want from the documents, such as competitive analysis, due diligence or proposal preparation.
- Claude creates a summary file, in the same folder as the PDFs unless you name another location.
- Claude reads one PDF, writes what matters to the summary file, and moves on to the next.
- Claude adds a section of insights that run across all the documents.
- Claude gives you the finished summary.
Claude writes the findings to disk after each document, so the earlier PDFs can leave Claude’s context and the findings stay safe. Claude also reports any PDF that is unreadable, image-only or damaged.
Filling Forms
Section titled “Filling Forms”The skill first checks whether the PDF has fillable form fields.
- Fillable form. Claude extracts the list of fields and fills them with your data.
- Form without fields. Claude converts each page to an image, works out where each answer should go, checks those positions with validation images, and adds the answers as text annotations.
The scripts for both routes ship in .claude/skills/processing-pdfs/scripts/. The completed form is saved next to the original.
What You Get
Section titled “What You Get”| Output | Format |
|---|---|
| Extracted text | Shown in the conversation, or saved as .txt or .md |
| Extracted tables | Markdown tables, or saved as .csv or .xlsx |
| Merged or split PDFs | Saved next to the source, or where you specify |
| Bulk summaries | A markdown summary file in the folder of PDFs |
| Filled forms | A completed PDF next to the original |
The workspace .gitignore excludes PDFs and CSV files, so they are not committed to git. Keep source PDFs on your cloud storage and give Claude their path.
Troubleshooting
Section titled “Troubleshooting”| Problem | Cause | What to do |
|---|---|---|
| The PDF returns no text | It is a scanned image, not text | Ask Claude to convert the pages to images and run OCR |
| Claude runs out of context with several PDFs | Large files read one after another | Ask for the Extract-Summarise-Release workflow |
| “File too large” | The PDF is over the single-read limit | Give Claude the file path so it can check the scale and read in chunks |
| Tables come out garbled | Merged cells or a complex layout | Ask Claude to adjust the table settings, or extract the text and rebuild the table |
Customising the Skill
Section titled “Customising the Skill”The skill file and its three reference files are customisable engine files, so your edits survive /update through a three-way merge. The reference files are workflows.md for code examples and the large-file workflows, reference.md for further libraries, and forms.md for form filling.