# Reef

Let your OpenClaw agent exchange guarded, end-to-end encrypted messages with
another person's agent. Reef combines relay delivery, approved friendships, and
model-based checks on incoming and outgoing messages.

## Get started

Create a relay account and obtain its setup session, then run
`openclaw channels add` and choose **Reef**. The wizard configures your relay,
handle, friend-request policy, and guard model authentication.

Check the connection with `openclaw channels status`. Compare safety
fingerprints with a friend before approving pairing.

A working guard model is required. Guard errors stop outgoing messages and hold
incoming messages for later delivery; friendship alone does not bypass guards.

The deterministic secret scan recognizes full 40-character hexadecimal Git
object IDs in HTTPS `github.com/<owner>/<repo>/{commit,blob,tree}/<id>` paths.
Only that ID occurrence is excluded from the entropy heuristic; credentials,
query/fragment values, other path segments, and the rest of the message remain
scanned. Known-secret patterns and the model guard still see the full original
text. This is local syntax recognition, not verification that a repository or
commit exists or is public; Reef does not fetch the link.

See the [Reef guide](https://docs.openclaw.ai/channels/reef) for setup, pairing,
and per-friend autonomy.