Security fixes are provided for the latest published release only. Older versions do not receive backports; update to the latest release to receive fixes. Before the first publication, fixes target the current source branch.

Use Report a vulnerability on ruban-24/switchboard's Security page to send a private report. If that option is unavailable, open an issue requesting a private contact channel without including exploit details, keys, or private data. There is no guaranteed response or resolution time.

Include affected versions, a minimal synthetic reproduction, and the impact. Do not post API keys, OAuth links, native credentials, full transcripts, or private repository content in public issues.

The local proxy forwards native inference to the tool's provider. Extracted task text is also sent to the selected Jev provider for classification, even when local raw-prompt capture is disabled. Native CLI transcripts and provider data retention are separate from Switchboard's local history settings.

Interactive switchboard init saves classifier credentials in connection.json beside the personal policy, with owner-only read/write permissions. The key is plaintext, so protect that file and its backups. Environment overrides are also supported; only the source-checkout commands load .env.local. Keep credentials out of policy.json and Git. Classifier keys are removed from the native child environment. The proxy binds to loopback with a per-launch secret; this is not intended to protect against other code already running as the same local user.

Before sharing diagnostics, inspect and redact them. See privacy and data flow for recipients and storage, and the native CLI guide for supported execution modes and current limitations.