Security fixes are applied to the latest release.

Do not open a public issue for a vulnerability that could expose prompts, credentials, or host files. Use GitHub's private vulnerability reporting feature on the repository's Security tab. Include the affected version, reproduction steps, impact, and any suggested mitigation.

fablex launches the locally installed Claude Code and Codex CLIs with the user's existing authentication. It does not collect telemetry, maintain its own prompt database, or transmit credentials itself. Prompts are sent to the selected model providers by their respective CLIs. Claude Code may persist its resumable session under its normal configuration; Codex handoffs use --ephemeral.

Responses cross a model trust boundary. Prior-agent output is bounded, JSON-encoded, and labeled as untrusted reference data before it is passed to the other agent. This reduces accidental instruction following but cannot guarantee that a model will ignore adversarial text quoted from a repository. Review repository instructions and agent changes as you would when using either provider directly.

Codex defaults to workspace-write and requires a Git repository. Treat --allow-non-git, danger-full-access, extra CLI arguments, and third-party modifications to this package as elevated-trust choices. The uninstaller removes only a launcher that resolves to the installed bridge and a plugin directory whose manifest identifies this package.