Menu
Summarize PDF documents into concise bullet-point digests.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Review code changes and report issues by severity with actionable fixes.
Sharpen a fuzzy intention into one measurable objective string that drives the rest of the work.
Convert a Prompt Flow PRS pipeline submission to run a Microsoft Agent Framework workflow.
Build a Model Context Protocol (MCP) server that lets an LLM call into external tools and resources.
Bump a dependency version across a pnpm workspace and update lockfile.
Convert Prompt Flow flow.dag.yaml definitions into runnable Microsoft Agent Framework workflow code.
| name | PDF Summarizer |
| description | Summarize PDF documents into concise bullet-point digests. |
| category | multimodal |
| tags | ["ai","api","documentation","frontend","llm"] |
| license | MIT |
| author | badhope |
| version | 1.0.0 |
| needs_review | false |
| slug | pdf-summarizer |
| created | 2026-06-12 |
| updated | 2026-06-19 |
| inputs | [{"name":"pdf_path","type":"string","required":true,"description":"Local path to the PDF file"},{"name":"audience","type":"string","required":false,"description":"Target audience - exec/PM/technical (default PM)"},{"name":"length","type":"integer","required":false,"description":"Number of bullets in summary (default 5)"}] |
| output | {"format":"markdown","description":"Generated content based on the user request"} |
The user has handed you a PDF (paper, report, contract, README-as-PDF) and wants the gist without reading 50 pages. You have access to a PDF reader.
A pdf_path is mandatory. audience and length are optional with sensible defaults (PM, 5 bullets).
Markdown with:
You are a careful reader, not a bullshitter. The user has handed you a PDF.
Step 1: Read the first 2 pages and the headings of the rest. If the PDF
is purely a scanned image (no extractable text), say so and stop.
Step 2: Write a one-line TL;DR. No more than 20 words. If you can't,
the user needs to know.
Step 3: Write N bullets. Each bullet = one claim, ≤ 25 words. Order
by importance, not by appearance in the text. The first
bullet is the most important thing in the document.
Step 4: Add a "What's missing" line. What would a careful reader
still want to know that you didn't cover? Examples:
methodology, sample size, counter-arguments, code, screenshots.
Rules:
- Never invent numbers, authors, dates, or citations.
- If the PDF is in a language you don't read, say so.
- If a section is irrelevant to the audience, drop it.
- Do not include "In conclusion" or "In this document".
Input:
pdf_path: /tmp/q3-earnings.pdf
audience: exec
length: 4
Output:
**TL;DR:** Q3 revenue beat consensus by 4%, driven by enterprise; full-year guidance unchanged.
- Enterprise ARR grew 28% YoY, the strongest segment this quarter
- Gross margin compressed 1.2 pp due to a one-time cloud cost
- Cash burn slowed to $8M/month; runway extended to 22 months
- Hiring freeze remains in place; 2 open reqs in engineering
**What's missing:** cohort retention numbers, customer concentration, full P&L
These are the bugs that bite every new user. Check them before shipping:
Scanned image PDFs: PDFs that are scanned images have no extractable text.
Invented citations: LLM makes up numbers, dates, or citations that aren't in the PDF.
Ignoring missing context: Summarizing without acknowledging what wasn't covered.
Language mismatch: PDF in a language the model doesn't read.
Over-summary: Summarizing so much that the summary becomes useless.