Only the latest minor release receives security fixes.

Do not file a public GitHub issue for security vulnerabilities.

If you find a security issue (such as a way to escape the destructive-command guard, leak the API key, inject untrusted content into Gemini prompts, or bypass the SubagentStop verdict handler), please report it privately:

• Email: through the repository owner's GitHub profile
• GitHub Security Advisory: open a private security advisory

Include:

• A clear description of the vulnerability
• Reproduction steps or a proof of concept
• The version affected
• Any suggested mitigation

• Acknowledgement: within 7 days
• Initial assessment: within 14 days
• Fix or mitigation plan: within 30 days for high-severity issues

In scope:

• Hook scripts in hooks/ (any way to bypass blocking, leak data, or inject untrusted content)
• The userConfig API key flow
• Subagent definitions in agents/ (prompt injection, tool misuse)
• The MCP server registration in plugin.json

Out of scope:

• Vulnerabilities in gemini-mcp itself (report there)
• Vulnerabilities in Claude Code, MCP, or Google AI Studio (report to those vendors)
• Issues that require local code execution to exploit (the user already trusts their own machine)