The latest published version on npm is the only one that gets fixes.

Please don't open a public issue for a security problem.

Use GitHub's private vulnerability reporting instead. Expect a first response within a week.

Please include what you found, how to reproduce it, and what an attacker gets out of it.

Worth knowing if you're assessing risk — the surface is small but not empty:

• Your API key is read from the GEMINI_API_KEY (or GOOGLE_API_KEY) environment variable. It is sent to Google over HTTPS in the x-goog-api-key header, and it is never written to tasks.db, never logged, and never printed — including in error output.
• Your prompts and results are stored unencrypted in SQLite at ~/.gemcatch/tasks.db (or GEMCATCH_HOME). Anyone who can read that file can read your research. It's an ordinary file with ordinary permissions — treat it like your shell history.
• GEMCATCH_BASE_URL redirects every API call, key included. It exists for proxies, gateways and the test suite. Don't point it at a host you don't trust, and be aware that anything setting it in your environment can capture your key.
• Network egress is limited to generativelanguage.googleapis.com, unless you override GEMCATCH_BASE_URL.

In scope: anything that leaks the API key, leaks the task store to another local user, or lets a crafted API response cause code execution.

Out of scope: what the model itself says. Prompt injection and model output are Google's domain, not this tool's — gemcatch prints the response it is handed.