Docs
Security and disclosure.
Regente is a local-first coordination layer. It helps participating agents avoid accidental collisions, route handoffs, and preserve an auditable local event trail. It is not an operating-system sandbox or a hard boundary against arbitrary local processes.
Report privately
Send suspected vulnerabilities to security@regente.dev. Avoid public issues for reports that could help someone bypass claims, forge attribution, leak tokens, or tamper with release/update flows.
Include the affected version or commit, reproduction steps, expected versus observed behavior, and any logs with secrets removed.
Security scope
hooksPre-edit and session hooks should fail predictably and avoid stale packaged paths.MCP bridgeAgent tools should preserve identity, validate inputs, and avoid unsafe local writes.local serverHTTP/WebSocket actions should bind to loopback by default and require local authorization.claimsFences, releases, stale sweeps, and handoffs should not allow accidental ownership theft.signingSigned events should reject forgery and keep old key history verifiable after rotation.releasePackaging, notarization, checksums, update metadata, and installer redirects should stay verifiable.Local-first boundaries
Workspace state lives in .regente/. Optional integrations such as GitHub sync, Slack or
Discord alerts, and planner model calls are operator-configured. Regente should never require sending
source code to a Regente-hosted service for basic coordination.
The honest claim is: Regente prevents accidental coordination collisions among participating agents. It does not prevent a local process from editing files outside Regente, disabling hooks, or using an operator escape hatch.
Handling secrets
Do not paste live API keys, GitHub tokens, Apple signing credentials, private keys, webhook URLs, or customer code into public issues or discussions. Use placeholders in reproductions and rotate any credential that may have been exposed.