There's no tagged release yet. main is what's supported — if you're running crate-wiki, uv tool upgrade crate-wiki puts you on it.

Report it privately through GitHub: open a draft security advisory. That reaches me without the report being public first.

If that form isn't available to you, open a normal issue saying only that you have a security report and how to reach you — no details in the issue — and I'll follow up privately.

I'm one person doing this alongside other work, so I can't promise a response time. I will acknowledge what I receive and say plainly whether I'm going to fix it.

crate-wiki reads local AI-assistant session transcripts and writes local Markdown. The engine makes no network calls of its own and has no server and no telemetry — the one operation that reaches outward is /lint, which may ask your assistant to run a web search. So the interesting failures are all about data staying where it should:

• Anything that causes vault content, session transcripts, or file paths to leave the machine.
• Anything that writes content from a private raw section into wiki/, which is committed (ADR-0006 is why crate lint checks for exactly that).
• Anything that lets a crafted transcript file cause code execution, or a write outside the vault, when it's parsed.
• Anything that breaks the isolation between a work vault and a personal one (ADR-0001).

Out of scope: the security of the AI assistants whose transcripts this reads, and anything you deliberately put in a vault and then pushed to a remote you chose.

Please don't include real transcript content, vault pages, client names, or internal paths in the report — describe the shape of the problem, or build a synthetic file that reproduces it. This repo's history is public and permanent, and so is anything that ends up in an advisory.