My fact-check skill checks a Google Doc or web page, generates a review interface, then writes the fixes I approve back into the doc.

I run:

The agent extracts the text and lists every checkable claim—dates, figures, quotes, who did what, the representation of linked content—and verifies them with parallel subagents. A separate pass hunts for overstatements. Every quote must be re-found by string search in a saved copy of the source, so the agent can’t invent one.

The review interface:

There’s a source viewer to see the quoted passage in context:

I review the outputs, then click “Send to agent”, and the agent writes them back into my Google Doc:

N.B. I made this in ~1 hour, and I’ve only used it a few times. I’ve been very happy with the results, but the fact checker prompts and agent orchestration could likely be improved a bunch.