# Security policy

Hound is a security product, and its agent runs as a root daemon that can
block execution. That earns a low bar for reporting problems and a high bar
for how we handle them.

## Reporting a vulnerability

Email **security@houndav.com** with enough detail to reproduce. Please report
privately and give us a chance to ship a fix before public disclosure. You
will get an acknowledgement, and we will keep you updated as we work it.

In scope:

- The agent (`houndd`, `hound`, `hound-gui`, `hound-mcp`) — memory safety,
  privilege escalation, the execution gate, the fanotify path, the MCP server.
- The definitions pipeline — signing, verification, the update client, pack
  and licence parsing.
- The download and definitions hosts, and this project's release artifacts.

Especially interesting: anything that lets an unsigned or tampered definition
pack, rules pack, or release manifest be accepted; any way to make the agent
quarantine or block something it should not; any path traversal or RCE reachable
through a pack, a licence token, or the update flow.

## What we commit to

- Definitions, rules packs, and release manifests are Ed25519-signed and
  verified **before** parsing, against a key whose public half is compiled into
  the agent. Verification failing degrades to a safe state, never to a broken
  one.
- No telemetry: the agent's only network call is fetching signed packs, and
  that can be pointed at your own mirror.
- We will credit reporters who want it, and we will not pursue good-faith
  research conducted under this policy.

## Verifying a report against the source

The agent is Apache-2.0 and public at
<https://github.com/HoundAV/hound-antivirus>. You are being asked to trust a
root daemon; you should be able to read exactly what it does.