CIB is pre-1.0. Security fixes target the latest released minor version.

Use GitHub's private vulnerability-reporting form from the repository Security tab. Do not open a public issue for suspected credential exposure, command injection, sandbox escape, unsafe fixture handling, or evidence disclosure.

Include the affected version, reproduction steps, impact, and any suggested mitigation. The maintainer will acknowledge a complete report within seven days and will coordinate disclosure after a fix is available. This is a volunteer project, so the policy is a target rather than a paid support guarantee.

CIB invokes coding agents and local commands. Treat benchmark fixtures and provider output as untrusted data.

• Scientific runs use read-only agent sandboxes and approval_policy: never.
• Each trial receives a unique fixture, HOME, CODEX_HOME, nonce, and session.
• The default Promptfoo profile disables cache, sharing, and thread persistence.
• CIB links existing Codex auth into isolated trial homes; it never copies auth contents into public manifests.
• Generated results/ may contain local paths, model output, synthetic nonces, and auth symlinks. Never publish it without a separate sanitization review.
• Run only benchmark cases and adapters you have inspected and authorized.

The deterministic publication check and GitHub secret scanning reduce leakage risk, but neither replaces review of the exact files and Git history being published.