Report it privately through GitHub's private vulnerability reporting: report a vulnerability (the repository's Security tab, then Report a vulnerability). Please do not open a public issue, pull request or discussion for it.

Include what you can of:

• what the problem is and what an attacker gains (another org's data, a secret, a gate that does not hold);
• the version or commit, and whether it is a self-hosted Commissary or Commissary Cloud;
• steps to reproduce, or a proof of concept;
• any logs or requests that help, with keys and tokens removed.

We read every report, reply on the advisory, and keep you posted while we confirm and fix it. Once a fix is deployed, we publish the advisory and credit you if you wish. This is a small project: there is no bounty and no fixed response time, but a report that shows a leak of secrets or one org reaching another's data comes first.

Only main and Commissary Cloud, which deploys from main, get fixes. There are no release branches. A self-hosted Commissary updates by deploying main (README).

docs/THREAT-MODEL.md says what Commissary protects, who it trusts, how it limits what a steered agent can do (ADR 0044), and the risks it accepts. A report that shows one of its defences not holding is in scope; behaviour the threat model lists under "Accepted risks" is by design.