Security policy

Reporting a problem

Please report security problems privately, not in a public issue: use Report a vulnerability on the repository's Security tab. If you can't, email [email protected]. Include what an attacker could do, the steps or a message that shows it, and the version or commit you tested.

satchel.email is maintained by one person, so replies are best effort, usually within a week. You'll hear when the report is confirmed and when a fix ships; tell me if you'd like credit in the release notes.

What's covered

satchel.email reads untrusted mail and lets AI agents in, so these matter most:

• a message, invite or attachment that runs script, reaches the network, crashes or hangs the app, or corrupts the local store;
• an agent (through MCP or the ~/Satchel/ folder) doing more than read and propose: sending, deleting, changing settings or its own policy;
• reading or writing files outside the data directory and the AI folder;
• leaking stored credentials or OAuth tokens;
• satchel serve or the desktop app accepting requests from other sites or other users on the machine.

Decision 0032 describes the local threat model. Fixes go into the latest release only.

Please don't

Test against mail accounts that aren't yours. scripts/dev-servers reset starts a local mail server with test accounts for this.

This page matches SECURITY.md in the repository. Security tools can also find this policy at /.well-known/security.txt.