OwnerResolutionCheck

What it does

Refuses any request that doesn't have a resolvable commit owner. This is the load-bearing check — signet's whole architectural premise is that every action attributable to someone. A request without an owner is a request whose audit row reads "unknown caller did unknown thing," which defeats the audit chain's purpose.

Stage

ADMISSION — runs before the request is forwarded upstream.

How owner resolution works

The check tries headers in this precedence:

1. X-Commit-Owner: human:<principal> → Owner.human(principal)
2. X-Agent-Id: agent:<id> → Owner.agent(id) (the agent: prefix is required; bare values are rejected)
3. X-Policy-Name: <name> (+ optional X-Policy-Version) → Owner.policy(name@version)
4. Already-resolved owner on the context (e.g. set by LoopbackTrustCheck)

If all four miss:

• require_owner=True (default): block with reason "no commit owner could be resolved". The proxy returns HTTP 403.
• require_owner=False: fall back to Owner.policy("unattributed") and audit a warning. Use only during enforcement shakedowns.

Configuration

from signet.checks import OwnerResolutionCheck

# Strict (recommended; default)
OwnerResolutionCheck()

# Permissive — for traffic-pattern observation before turning enforcement on
OwnerResolutionCheck(require_owner=False)

Header conventions

Header	Example	Owner type
X-Commit-Owner	human:alice@example.com	human
X-Agent-Id	agent:nightly-syncer	agent (requires the agent: prefix)
X-Policy-Name	acme.security	policy
X-Policy-Version	v3	(combined with name)

Internal callers

If your services are co-located with signet (Rolling Memory, Smart Router, MCP, internal tools), put LoopbackTrustCheck before OwnerResolutionCheck in the pipeline. The loopback check auto-resolves owner for trusted IP ranges (loopback + Tailscale CGNAT) so internal traffic doesn't need to know about the header convention.

Audit row example

When this check blocks:

{
  "owner_type": "unresolved",
  "owner_id": "",
  "approval_chain": [],
  "check_name": "owner_resolution",
  "decision": "block",
  "reason": "no commit owner could be resolved",
  "metadata": {
    "_check_name": "owner_resolution",
    "_stage": "admission",
    "hint": "set X-Commit-Owner: human:<principal>, ..."
  }
}

When it allows:

{
  "owner_type": "human",
  "owner_id": "alice@example.com",
  "approval_chain": ["human:alice@example.com"],
  "check_name": "owner_resolution",
  "decision": "allow",
  "reason": "owner resolved: human:alice@example.com",
  "metadata": {
    "source": "human:alice@example.com"
  }
}

Common bypass attempts (and why they fail)

Attempt	Result
Omit the header entirely	Decision.BLOCK (no header → no resolution)
Send X-Commit-Owner: alice (missing prefix)	Decision.BLOCK (precedence requires human:)
Send X-Commit-Owner: ""	Decision.BLOCK (empty value)
Spoof X-Commit-Owner: human:admin	Allowed — but the audit row records "admin" as the owner. Spoofing doesn't help an attacker because they're now on the hook for the audit. The check is about attribution, not identity verification. Pair with auth at the network or platform layer.

This last point is important: signet's owner resolution is attribution, not authentication. A platform-layer auth gate (mTLS, OIDC, IAM) decides whether the caller's claim of being human:alice is true. signet's job is to refuse anything where there's no claim at all.