foolish-bandit makes Gary admit when it can't show its work

Every AI answer now wears its verification status on its sleeve - cited, or flagged as unchecked.

chat-uicompliance

foolish-bandit reworked the interface around an existing feature: answers that point back to a source document. When an answer carries citations, it now sits inside a quiet reminder to check those sources before relying on them. When an answer has no citations at all, it gets an amber warning that says, plainly, to treat the answer as unverified.

The same framing runs through the review, draft, and explain screens. The review page tells you findings can only cite a source when a document is selected. The explain page warns that text pasted in raw won't produce citations. The draft page now says, in two places, that drafts must be reviewed before use. Nothing under the hood changed - this is entirely about setting the user's expectations at the moment they're reading an answer, rather than burying it in a generic disclaimer.

So what Anyone deploying legal AI to non-lawyers should study this: it turns "always verify" from fine print into a per-answer signal.

View this fork on GitHub →

Spotted something wrong? Or know the PR text has fresher detail than the writeup above?

Commits in this thread

1 commit from foolish-bandit/gary, oldest first. Source extracted verbatim from the harvested git log.

SHA Subject Author Date
b178b7ed polish cited answer expectations SONOMOS 2026-05-07 ↗ GitHub

Capture this thread into my fork

Download a single Markdown prompt that tells Claude how to port every commit above into your working tree — adapting paths and structure to match your repo. Run it via claude -p < capture-thread-264.md from inside the repo you want the changes in.

⬇ Download capture-thread-264.md