Skip to content

brycewang-stanford/Auto-Empirical-Research-Skills / skills/13-scunning1975-MixtapeTools/skills/split-pdf

split-pdf

Verified

Download, split, and deeply read academic PDFs. Use when asked to read, review, or summarize an academic paper. Splits PDFs into 4-page chunks, reads them in small batches, and produces structured reading notes — avoiding context window crashes and shallow comprehension.

FileKindSize
SKILL.mdskill md6.5 KB
methodology.mdmarkdown4.4 KB

methodology.md

View on GitHub
# Why PDF Splitting Works: The Methodology

This document explains the reasoning behind the split-pdf skill. It is reference material for humans — the SKILL.md file contains the actual instructions Claude follows.

---

## The Problem

Claude Code (currently Claude Opus 4.5, model ID `claude-opus-4-5-20251101`) can read PDFs, and it has a very large context window (200k tokens in). In principle, a 40-page academic paper should fit comfortably. In practice, it doesn't work well. Claude Code regularly chokes when asked to read long PDFs, and this manifests in two distinct ways:

**Problem 1: Session-breaking "prompt too long" errors.** PDF rendering into tokens is expensive. PDFs are not plain text — they are containers for fonts, vector graphics, embedded images, multi-column layouts, mathematical notation, tables, and footnotes. When Claude Code ingests a PDF, it must convert this complex layout into a linear token stream. A long PDF can produce a token sequence that, combined with the rest of the conversation context, exceeds the model's input limit. When this happens, Claude Code returns a "prompt too long" error. There is no way to recover — the session is broken, and all context is lost unless it has been externalized to files.

**Problem 2: Shallow, unreliable reading.** Even when the PDF does not trigger a hard failure, comprehension degrades badly for long documents. The model's attention over many tokens from a single dense document is not uniform — it attends more carefully to the beginning and end, and less carefully to the middle. The result is that Claude "skims" rather than "reads." It catches the abstract and introduction, gets fuzzy on the methodology, and often misses or fabricates details from the results and appendix. It truncates content silently, hallucinates details it didn't actually parse, or produces shallow summaries that miss critical methodological details.

These are related but distinct problems. The first is a hard constraint — the session dies. The second is a soft degradation — the session continues but the output is unreliable. Splitting addresses both.

## Why Batched Reading Works

Reading 3 splits at a time (~12 pages) does several things:

1. **Forces the model to focus.** With only 12 pages of content, Claude cannot skim. It has to engage with the material at a granular level — the specific equations, the exact data sources, the precise variable definitions.

2. **Creates natural checkpoints.** After each batch, the model writes down what it has learned so far. This means its understanding is externalized into a markdown file. If it makes an error in batch 1, you can catch it before batch 2 builds on it.

3. **Accumulates rather than summarizes.** When you ask Claude to read a full paper at once, it produces a summary — a lossy compression. When you ask it to read in batches and update running notes, it accumulates detail. The final notes are richer than any one-shot summary could be.

4. **Controls for the "front-loading" problem.** Claude's attention is not uniform over a long document. It attends more carefully to the beginning and end. By splitting the paper, every section gets to be "the beginning" of some chunk.

## Why 4-Page Chunks

Four pages is a sweet spot:
- Small enough that Claude attends carefully to every detail
- Large enough that logical sections (a methodology subsection, a results table with discussion) stay together
- Produces a manageable number of chunks (a 40-page paper = 10 chunks = 4 rounds of reading)

## Why the Pause-and-Confirm Protocol

The human pause between batches serves multiple purposes:
- **Review intermediate output** — catch errors before they compound
- **Redirect the reading** — ask follow-up questions, skip sections, or change focus
- **Prevent context drift** — Claude doesn't lose track of where it is in a long session
- **Control pacing** — some papers require more careful reading in specific sections

## Limitations

- **It is slow.** A 37-page paper split into 10 chunks, read 3 at a time, requires 4 rounds. Each round involves the user confirming "yes, continue." This is a 10-15 minute process rather than a 1-minute process.
- **Notes can become repetitive** if the paper revisits themes. Some manual editing of the final notes may be useful.
- **Assumes the paper is worth reading carefully.** For triage — quickly deciding whether a paper is relevant — reading just the first split (pages 1-4, which usually contains the abstract and introduction) is sufficient.
canvas-designanthropics/skillsCreate beautiful visual art in .png and .pdf documents using design philosophy. You should use this skill when the user asks to create a poster, piece of art, design, or other static piece. Create original visual designs, never copying existing artists' work to avoid copyright violations.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
docxanthropics/skillsUse this skill whenever the user wants to create, read, edit, or manipulate Word documents (.docx files) or Word templates (.dotx files). Triggers include: any mention of 'Word doc', 'word document', '.docx', '.dotx', or requests to produce professional documents with formatting like tables of contents, headings, page numbers, or letterheads. Also use when extracting or reorganizing content from .docx or .dotx files, inserting or replacing images in documents, performing find-and-replace in Word files, working with tracked changes or comments, or converting content into a polished Word document. If the user asks for a 'report', 'memo', 'letter', 'template', or similar deliverable as a Word or .docx file, use this skill. Do NOT use for PDFs, spreadsheets, Google Docs, or general coding tasks unrelated to document generation.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
pdfanthropics/skillsUse this skill whenever the user wants to do anything with PDF files. This includes reading or extracting text/tables from PDFs, combining or merging multiple PDFs into one, splitting PDFs apart, rotating pages, adding watermarks, creating new PDFs, filling PDF forms, encrypting/decrypting PDFs, extracting images, and OCR on scanned PDFs to make them searchable. If the user mentions a .pdf file or asks to produce one, use this skill.Stars: 198kUpdated Claude CodeCodexVerifiedInstall
pptxanthropics/skillsUse this skill any time a .pptx or .potx file is involved in any way — as input, output, or both. This includes: creating slide decks, pitch decks, or presentations; reading, parsing, or extracting text from any .pptx or .potx file (even if the extracted content will be used elsewhere, like in an email or summary); editing, modifying, or updating existing presentations; combining or splitting slide files; working with templates (.potx), layouts, speaker notes, or comments. Trigger whenever the user mentions "deck," "slides," "presentation," or references a .pptx or .potx filename, regardless of what they plan to do with the content afterward. If a .pptx or .potx file needs to be opened, created, or touched, use this skill.Stars: 198kUpdated Claude CodeCodexVerifiedInstall