Hermes holds credentials for production databases. Treat anything that touches secrets, subprocesses or logs as security-relevant.

The project is pre-1.0: only the latest release receives fixes. Once 1.0 ships, this section will name the supported line explicitly.

Report privately through GitHub: open the repository's Security tab and use Report a vulnerability. Please do not open a public issue for a vulnerability, and do not include a working exploit in the first report.

Useful in a report: what an attacker gains, the smallest reproduction you have, the affected version or commit, and your platform. You can expect an acknowledgement within a few days and a fix or a plan before any public disclosure.

These are commitments, so a violation of any of them is a vulnerability worth reporting:

• Secrets never touch disk in plain text. Passwords live in the operating system keychain — Keychain on macOS, DPAPI on Windows, Secret Service on Linux. Connection files are meant to be committed to a repository and contain no credentials.
• Secrets never appear in argv. Subprocesses such as pg_dump receive credentials through the environment or through a temporary .pgpass created with mode 0600 and removed afterwards, including when the process is killed.
• Secrets never appear in logs. Log output, exported reports and error messages are redacted.
• No network calls beyond your databases. No telemetry, no update pings and no AI services unless you turn them on yourself.
• No credential crosses into the frontend. The interface receives connection state, never a password.

Signing keys and their passwords exist only as CI secrets. The workflow that signs and publishes runs separately from the workflow that builds pull requests, so a pull request from a fork never has access to them. Every release publishes checksums.

In scope: credential handling, subprocess execution, log redaction, the update mechanism, and anything that lets a connection marked read-only or production perform a write it should not.

Out of scope: vulnerabilities in PostgreSQL itself, in the PostgreSQL client tools, or attacks that require an already-compromised operating system account.