Research & literature reviews
Skriptr can go and find the literature, not just work with what you already have. Ask it to research a topic and it searches academic databases and the web, tells you why each paper matched, and lets you decide what actually enters your library. Ask for a literature review and you get a matrix you can sort.
The rule that governs all of it: nothing lands in your library because the agent liked it. You pick.
How do I find sources I do not have yet?
Ask. "Find me recent work on incubation and creative problem solving" is enough to start.
What comes back is a set of result cards, each with the paper and the reason it matched, not a wall of links. From there you can import the ones you want, ask for more like a particular paper, or send the agent back with a narrower brief. Anything you import is indexed like any other source, so you can cite it and search inside it straight away.
The Research Companion is the version of this worth knowing about. Instead of running one search and handing back a list, it scopes the question with you first, searches, then spars with you about what came back: which paper is actually load-bearing, which result is thin, what the field calls this thing when it isn’t using your words. Searching an unfamiliar field is mostly a vocabulary problem, and that is the part it helps with.
How do I run a literature review?
Say so. "Do a literature review on X" starts it.
Before running anything, the agent checks two things: the scope it thinks you mean, and how you want the result delivered. You can have it as
- a spreadsheet matrix in your library, sortable and exportable,
- a summary in the chat, the main themes and the papers that matter, or
- a drafted section in your document.
Pick more than one if you want more than one. The confirmation step exists because a literature review is the most expensive thing you can ask for, and the worst outcome is a thorough review of slightly the wrong question.
What is in the literature matrix?
A row per paper and a column per thing you’re comparing: title, year, method, study type, whatever the review called for. It arrives as a spreadsheet in your library, so you can sort it by year, filter by method, or export it into the appendix of your thesis.
This is the artefact people underestimate. A matrix makes gaps visible in a way prose can’t. Sort by year and a dead decade jumps out. Sort by method and you find that every study pointing one way used the same design.
Can the agent argue with me about what it found?
Yes, and it’s one of the more useful things it does.
Devil’s Advocate takes a claim you’ve made or found and argues both sides, using evidence from your own sources. You read the two cases and decide which of your claims are actually weak. Only the ones you flag get strengthened, so it can’t quietly rewrite an argument you were happy with.
Structured Argument Building works the other direction: it analyses the claims in a passage, tells you which ones are weak or improvable and why, and strengthens only what you choose.
Both are there for the same reason. A supervisor who only ever agreed with you would be useless, and so would a research assistant who only ever confirmed your priors.
Where do I go next?
- Sources & your library. Where everything you import ends up.
- Citations & references. Citing what you found, in APA 7, IEEE or Chicago.
- Studying & learning. Understanding the field you just mapped.
Still stuck?
If a search is coming back with nothing useful, tell us what you asked and what you expected at contact. Bad results are usually a fixable prompt-and-scope problem, and we’d like to know which ones.