Ownward is currently in alpha. Security fixes are applied to the latest version on the default branch; older commits and forks are not maintained.

Please do not open a public Issue for vulnerabilities, credentials, private logs, or reproducible exploit details.

Use GitHub Private Vulnerability Reporting on this repository. Include:

• the affected version or commit;
• the relevant configuration and platform;
• reproduction steps and impact;
• any suggested mitigation;
• sanitized logs only, with tokens, personal content, paths, and identifiers removed.

If private reporting is not enabled yet, open a public Issue containing no sensitive or exploitable details and ask the maintainer to provide a private reporting channel.

Ownward is a local-first automation tool, not a security boundary:

• the dashboard listens on localhost by default;
• remote access requires a token and should be placed behind Tailscale or a trusted TLS proxy;
• coding agents may execute commands and modify files with the current user's permissions;
• worktrees reduce accidental changes to a main checkout but are not sandboxes;
• content sent to configured AI providers or external integrations leaves the local machine;
• long-term memory promotion, routine publication, and self-evolution deployment retain human approval gates.

Never commit config.json, prompts/owner.md, data/, vault content, OAuth credentials, API tokens, or raw conversation transcripts.