Project
Responsible use
What Overfull is for, how it supports research integrity, AI-use disclosure, human review, and what it will not help with.
Overfull exists to make papers and theses more correct. It checks; it does not write. This page explains the principles behind that and what we ask of people who use it, with or without an AI agent.
What Overfull is for
- Catching mistakes before submission: build errors, broken references, inaccessible PDFs, venue rule violations.
- Giving AI agents facts they cannot know from training data, so their mechanical work on a LaTeX project can be verified.
- Making changes reviewable, through small commits and highlighted-changes PDFs.
What it is not for
- Generating research content, references or data. Overfull has no command that writes prose, and its agent instructions forbid inventing results, numbers, quotes or citations.
- Producing papers at scale for others to sign (paper mills), or helping anyone submit work that is not their own.
- Hiding AI use. The disclosure skill drafts statements from actual use and is written never to claim that no AI was used.
- Evading venue rules, for example by squeezing spacing to meet a page limit or leaving identifying details in an anonymous submission.
The author signs
Whoever submits the document is responsible for every sentence, figure, number
and reference in it. A passing overfull check means the document builds, its
references exist and match the indexes, and it meets the configured rules. It
does not mean:
- the cited work supports the claim it is attached to;
- the results are correct or reproducible;
- the text is original;
- the document meets every rule of the venue (only the ones you configured).
Human review
When an agent works on your paper:
- Ask it to work on a branch and open a pull request.
- Read the highlighted-changes PDF from
overfull diff, not only the summary. - Check every new reference against what you intended to cite.
- Check every claim against your notes and data.
- Merge only what you would have written yourself, or would sign as written.
Disclosure
Venues and universities have their own, changing policies on AI use. Read the
current policy for your venue and follow it. If an agent helped, say how. The
write-disclosure skill drafts a statement from the repository’s history; you
edit and approve it. See Venues.
Integrity of the checks
Agents are instructed not to weaken the configuration to make checks pass: no removing checks, no ignore-list entries, no edits to venue rules. If you see a pull request that does this, treat it as a red flag and ask why.
Reporting misuse or problems
If you find a way Overfull’s tools or instructions make it easier to fabricate
content or hide AI use, please open an issue or follow SECURITY.md for
sensitive reports.