Only the latest release gets security fixes.

The mod's code: everything under hooks/, scripts/ and .claude-plugin/.

The part most worth a close look is the memory shadow log. It is off by default. When memoryShadow is on, the mod sends quoted session text to the model provider and appends to a local log file. The README states exactly what that mode sends and stores, and what its redaction can miss: see Memory shadow log (experimental). The steer setting is in scope beside it: off by default, at shadow it appends to a local log file, and at on it also appends a fixed note to the conversation that the model reads: see Steering (experimental).

Claude Code itself is out of scope here.

Use GitHub's private vulnerability reporting: open the repository's Security tab and choose Report a vulnerability. Please do not open a public issue for a security problem.