Readme of agentui-ai/pdf-ops

jsPDF draws exactly where you tell it and never complains. Text wider than the page is not wrapped — it is drawn off the edge. Text past the last line is not moved to a new page — it is drawn into nothing. The file opens fine and the content is simply missing. The person who finds out is holding the printout.

CASE                             SKILL  TRAP   EXPECTED           FROM MEMORY
text-runs-off-the-page           ok     yes   {"measure":0}       {"measure":1}
content-past-the-last-line       ok     yes   {"measure":0}       {"measure":46}
blank-trailing-page              ok     yes   {"pages":3,"sectio… {"pages":4,"sections":3}
units-are-millimetres            ok     yes   {"measure":0}       {"measure":1}
default-font-size-is-16          ok     yes   {"measure":0}       {"measure":2}
unsupported-glyphs-render-as-ga  ok     yes   {"flagged":["found… {"flagged":[]}
content-after-a-table            ok     yes   {"overlaps":0}      {"overlaps":1}
image-aspect-ratio               ok     yes   {"distortion":0}    {"distortion":3}
page-numbers-need-the-total      ok     yes   {"stamps":["Page 1… {"stamps":["Page 1 of 1…

SKILL  9/9  the recipes in SKILL.md produce a correct document
TRAP   9/9  the from-memory version gets it wrong

content-past-the-last-line is the one to look at: 46 lines drawn into nothing, and not a single error.

Install

# Any of ~75 agents (Gemini CLI, opencode, aider, …)
npx skills add agentui-ai/pdf-ops --agent gemini-cli --global

# Cursor
git clone https://github.com/agentui-ai/pdf-ops.git ~/.cursor/plugins/local/pdf-ops

# Codex
codex plugin marketplace add agentui-ai/pdf-ops && codex plugin add pdf-ops@pdf-ops

# Claude Code
claude --plugin-dir ./pdf-ops

No account, no service, no platform. Read skills/pdf-ops/SKILL.md directly if you would rather not install anything.

The idea: a layout you can test

import { createDoc, Flow } from "./skills/pdf-ops/scripts/layout.mjs";
import { offPageDraws } from "./skills/pdf-ops/scripts/inspect.mjs";

const doc = createDoc();                    // mm, A4, explicit font size
const flow = new Flow(doc, { margin: 18 });

flow.heading("Quarterly report");
flow.text(longParagraph);                   // wraps AND paginates
flow.footerPageNumbers();                   // last — the total is known only now

if (offPageDraws(flow).length) throw new Error("content fell off the page");

Flow records every write with its measured box, so that last line turns "the PDF looks wrong" into a failing test. That is the whole trick.

Check a finished PDF

npm install
node skills/pdf-ops/scripts/inspect.mjs out.pdf

1 page(s)

PAGE   SIZE (mm)        SIZE (pt)        FORMAT
   1   210 x 297        595.3 x 841.9    A4 portrait

METADATA
  title     Demo invoice
  author    AgentUI
  subject   — not set

Run the benchmark

npm install && npm run bench

About a second. Deterministic, offline, free — no LLM, no network, no fixtures on disk. Each case builds a real PDF and judges it by measurement: off-page draws from the document's own metrics, page counts parsed back out of the finished bytes with pdf-lib, aspect ratios from the image itself.

Two ledgers, because either half alone lies:

SKILL — does the recipe in SKILL.md produce a correct document? Under 100% is a broken promise, and the run exits non-zero.
TRAP — does the from-memory version get it wrong? A case both sides pass is not a trap; the report says so instead of inflating the score.

npm run bench -- --json for CI.

What is in it

pdf-ops/
├── skills/pdf-ops/
│   ├── SKILL.md              # nine traps, layout, tables, fonts, images, iOS downloads
│   └── scripts/
│       ├── layout.mjs        # Flow: wrapping, pagination, glyph check, image fitting
│       ├── report.mjs        # a document with a table that paginates properly
│       └── inspect.mjs       # CLI on a finished PDF + offPageDraws() for your tests
└── benchmark/
    ├── cases.mjs             # nine cases, each naive vs skilled
    └── run.mjs               # the two ledgers

Reference implementation is jsPDF + jspdf-autotable, in Node or the browser. SKILL.md says plainly when jsPDF is the wrong tool — editing an existing PDF is pdf-lib, reading one is pdfjs-dist, and print-grade typography is a headless browser.

Also see

excel-ops — the same treatment for spreadsheets. oee-ops — the same for manufacturing OEE / TRS. label-ops — the same for ZPL labels, thermal receipts and barcode check digits. If the user wants a hosted app rather than a script, AgentUI does that and agentui-tools is the agent plugin for it. Everything here works without either.

License

MIT — see LICENSE.