Sample output · 1 of 3
A few products, tools or plans compared on the same handful of criteria, from pages somebody read: a PDF with the table, the numbered sources and the log of every page that was opened, and the same report as markdown.
The first task in a new project also sets up its computer (about 3 min).
Compare options side by side in a short report
Try this prompt
Use the Market Research skill to compare three e-ink note tablets for long writing sessions. Judge them on price, battery life and sync. Read the pages yourself and give me a PDF I can send, with the sources numbered and a line for every page you opened, including the one you could not read.
Two files for a handful of things judged on the same criteria: report.pdf, the document — a cover that counts what was found and what was not, the comparison table with the number of the source behind every value, the numbered sources with the addresses they were read on, and the log of every open — and report.md, the same report as plain text. Plus the page images and a QA report.
| Field | Required | Hint |
|---|---|---|
| topic | Yes | What the comparison is about, at most 200 characters; it is printed on the cover |
| items | Yes | Two to eight things to compare, at most 80 characters each; the order is the order of the rows |
| criteria | Yes | One to eight things to judge each item on, at most 80 characters each; the order is the order of the columns |
| language | Yes | The language the report is written in, at most 35 characters |
| report | No | The title on the cover (at most 120 characters) and an optional one-line subtitle (at most 200) |
| findings | Yes | One finding per cell a page backs: the item, the criterion, what the page said, and the keys of the sources behind it |
| notes | No | Up to eight readings of the comparison; every note names the sources it came from |
| sources | Yes | Three to twenty-four pages the report leans on, each with a key the findings cite it by and a title |
| fetchLog | Yes | Every page that was opened, cited or not: the host and path, the outcome, the status, and the reason when it failed |
You say
Use the Market Research skill to compare three e-ink note tablets for long writing sessions. Judge them on price, battery life and sync. Read the pages yourself and give me a PDF I can send, with the sources numbered and a line for every page you opened, including the one you could not read.
None
Node 20+; Python with pypdf; Chromium (playwright-core) for the .pdf and the preview; the pages are read by the agent's own web tools or by the guarded open script
None
topic (What the comparison is about, at most 200 characters; it is printed on the cover), items (Two to eight things to compare, at most 80 characters each; the order is the order of the rows), criteria (One to eight things to judge each item on, at most 80 characters each; the order is the order of the columns), language (The language the report is written in, at most 35 characters), findings (One finding per cell a page backs: the item, the criterion, what the page said, and the keys of the sources behind it), sources (Three to twenty-four pages the report leans on, each with a key the findings cite it by and a title), fetchLog (Every page that was opened, cited or not: the host and path, the outcome, the status, and the reason when it failed)
onlypro, claude-code, codex, cursor