Security
If you have found something, we would rather hear it from you than from someone else.
Last updated 24 August 2026
Open a private security advisory on the repository. If you cannot, open a normal issue that says only that you have found a security problem and how to reach you, with no details in the public thread.
Please give us a reasonable window to fix an issue before publishing it.
Anything that lets one account read or change another account’s data, anything that lets an unauthenticated request reach a protected route, and anything that causes Pepiros to attribute a quote to a source it did not come from.
That last one matters as much as the first two here. A grounding tool that can be made to point at the wrong sentence has failed at the only thing it claims to do.
Also in scope: anything that gets past upload validation itself -- a spoofed file type, an oversized payload, or a page count that slips through uncounted before it reaches the parser.
Sessions are signed HTTP-only cookies with a 7-day lifetime and no silent renewal, and are server-side revocable: signing out, or signing out everywhere, actually kills the session rather than just clearing a cookie the server would still honor. Protected routes are enforced in middleware, not only hidden from search engines.
An MCP token carries its own scope (read-only, or read and write), can be pinned to a single workspace rather than every workspace an account owns, and can be revoked independently of the account session that minted it.
The verifier and its thresholds -- the thing this page’s own “what is in scope” section leads with -- are covered by five dedicated test files (anchor location, fuzzy matching, numeric entailment, reverse audit, tier assignment), not just code review.
There is no bug bounty, no formal SLA, and no third-party audit. This is a two-person project, and pretending otherwise would be its own kind of security problem.