LoopbackTrustCheck

What it does

Auto-resolves the commit owner for requests originating from trusted internal IP ranges — loopback (127.0.0.0/8, ::1) and Tailscale CGNAT (100.64.0.0/10) by default. Lets internal services co-located with signet (Rolling Memory, an internal MCP, your own gateway) skip the X-Commit-Owner header dance for trusted traffic without disabling owner enforcement for external callers.

Useful pattern: stack LoopbackTrustCheck before OwnerResolutionCheck. Loopback traffic auto-resolves to policy:internal-loopback; OwnerResolutionCheck sees ctx.owner.is_resolved and passes through.

Stage

ADMISSION — runs before any forwarding decision.

Configuration

from signet.checks import LoopbackTrustCheck

# Defaults (loopback + Tailscale CGNAT)
LoopbackTrustCheck()

# Add additional trusted CIDRs (your own VPC range, Wireguard subnet, etc.)
LoopbackTrustCheck(extra_trusted_cidrs=("10.42.0.0/16", "fd7a:115c::/48"))

The default trusted set:

CIDR	Purpose
127.0.0.0/8	IPv4 loopback
::1/128	IPv6 loopback
100.64.0.0/10	RFC 6598 (Tailscale CGNAT range)

Owner assignment

Source IP	Resolved owner
127.0.0.1 (loopback)	policy:internal-loopback
100.x.y.z (Tailscale CGNAT)	policy:internal-tailnet:100.x.y.z
Your extra_trusted_cidrs	policy:internal-loopback (same as loopback)
Anything else	not resolved (next check runs)

Audit row example

When loopback traffic resolves:

{
  "owner_type": "policy",
  "owner_id": "internal-loopback",
  "approval_chain": ["policy:internal-loopback"],
  "check_name": "loopback_trust",
  "decision": "allow",
  "reason": "loopback IP resolved to policy:internal-loopback"
}

Caveat

This check trusts the source IP. If your reverse proxy doesn't strip / overwrite a forwarded-for header that signet might mistake for the real source, an external caller could spoof their way into the trusted range. Make sure your reverse proxy sets the source IP signet sees to the actual peer, not whatever the caller claims.