Security fixes are handled on the main branch. Until this project publishes stable releases, report issues against the latest commit on main.

Please report sensitive security issues through GitHub security advisories for this repository when available.

Do not include secrets, exploit details, private logs, or account data in public issues. If advisories are unavailable, open a minimal public issue that says a private security report is needed, without sensitive details.

In scope:

• Prompt or path handling that can expose Codex private runtime context without explicit user authorization.
• Environment leakage beyond the documented Grok child-process allowlist.
• Job-state handling that can escape the plugin-owned private state directory or bypass the authorized workspace boundary.
• Process-tree, cancellation, timeout, or prompt-cleanup failures on supported macOS/Linux hosts that can leave Grok work running after a terminal result.
• MCP tool behavior that misrepresents incomplete Grok output as a final result.

Out of scope:

• Vulnerabilities in the Grok CLI, xAI services, Node.js, npm, Codex, or the user's local operating system.
• User-provided prompts that intentionally include secrets or sensitive data.
• Social engineering or attacks against maintainers or users.

The plugin is designed to avoid copying Codex hidden context, system/developer messages, hidden reasoning, tool outputs, secrets, and private runtime paths into Grok prompts unless the user explicitly authorizes that risk.

Anything that can make the plugin report an incomplete or cancelled Grok run as a completed answer is treated as a security-class defect for release purposes: it goes public in the same session it is fixed. See the release cadence section in CONTRIBUTING.md for the process and for the 0.2.0/0.2.1 incident that set it.