Study a PDF with a quiz-based tutor

What it is

This skill turns the agent into a study partner built around one document. You hand it a PDF, such as a textbook chapter, a vendor manual, a standards document or a course handout, and it works from that file instead of general knowledge. It asks you questions, waits for your answer, then tells you whether you were right and which page supports the answer. It keeps a running record of the questions you got wrong so later sessions can return to those weak spots first.

The agent reads the file once, breaks it into passages, and holds them in a local index. Every reply it gives is tied back to a passage and a page number, so you can check it yourself. Nothing leaves your machine beyond what the agent already needs to run.

When to use it

Reach for this when you have a dense document and a deadline. Examples: a certification candidate working through a 300-page exam guide, a new hire learning an internal operations manual, a student revising for a closed-book test, or anyone who has read a report once and retained very little.

It suits material with structure: headings, numbered clauses, tables, worked examples. It is a poor fit for a single-page flyer with almost no content to question you on.

How to use it

1. Install the skill in your agent environment, following the install steps that ship with it. In most setups this means dropping the skill folder into your skills directory and confirming the agent lists it.
2. Start a session and give the agent the file path to your PDF, plus one line about your goal. For instance: "I sit the facilities safety exam in three weeks. Build questions from chapter 4 onward and start at medium difficulty."
3. Let it index the document. It will confirm how many pages it read and flag any it could not parse, such as scanned images with no text layer.
4. Answer the first question in your own words. Do not ask for the answer first; the point is to retrieve, not to recognise.
5. Read the feedback. It should name the page, quote the relevant line, and say what your answer missed.
6. Ask it to go deeper on anything you got wrong, for example: "Explain why option B fails, using the section on pressure relief."
7. At the end, request a summary of missed topics and a short set of follow-up questions for your next session.

What a good result looks like

Questions stay inside the document and match its vocabulary. Feedback cites a page you can flip to in seconds. The running list of weak topics is specific enough to act on, such as "confused two alarm thresholds" rather than "needs more revision". After two or three sessions you should notice the same topics coming back until you clear them, and the questions getting harder as your answers improve.

If the agent starts inventing content that is not in the file, tell it to quote the supporting line or say it cannot find one. That single instruction keeps the whole session honest.