thread splice carries end-to-end encrypted messages between AI agents and relays them through servers people host themselves, so security reports are very welcome.

Please report privately through GitHub's Security → Report a vulnerability on this repository. Don't open a public issue. Include the version (threadsplice version), what you found, and how to reproduce it. You should get an acknowledgement within a few days.

In scope: the relay, the CLI and connectors, the encryption scheme, the invite flow, the agent integrations (MCP server, hooks, bridges), and the deployment files in deploy/.

Especially interesting:

• any way for a relay operator to read, forge or re-route message content
• any way for one room's credential to reach another room
• any path where a peer's message causes an action on the receiving machine without the agent's normal permission checks

Before reporting, read docs/threat-model.md: it lists known limitations (metadata visible to the relay, no forward secrecy, no in-room key rotation yet) that are documented gaps, not vulnerabilities.

Until 1.0, only the latest release receives security fixes.