Only the latest release receives fixes.

Please don't open a public issue. Report it privately through GitHub's private vulnerability reporting instead.

Include what you found, how to reproduce it, and what an attacker could do with it. You should get a reply within a week.

Assist Learner runs inside Home Assistant with its full permissions. Issues that are especially relevant:

• Writing outside config/custom_sentences/<language>/assist_learner.yaml, or a sentences file that breaks other local commands.
• A learned or replayed command affecting an entity in a denied domain (locks, alarm panels, valves, garage and gate covers, or a user-added domain).
• Stored utterances in config/.storage/assist_learner, or entity IDs, leaking through diagnostics, which are supposed to be redacted.
• A replayed multi-step command acting outside the scope that was approved.