Governance state anchor for external decision receipts

[Liuyanfeng1234]

Governance state anchor for external decision receipts

Context

We operate a governance composition system built on three layers:

• L1 (Signature Core): CompositionRef — SHA-256 JCS (RFC 8785) to anchor action_ref + delegation_ref + revocation_ref into a single composition_ref hash.
• L2 (Verifier Policy): QUASAR — four-tier trust thresholds (A1/A2/A3) that map governance state to verifier posture.
• L3 (Profile Constraints): Signed boundary constraints (8 prohibitions + 5 constraint types) documented as a profile layer with RFC-form specification.

Alignment with HELM's external receipt model

HELM's external_decision_receipt.v1 records the execution decision (allow/deny with Ed25519 signature). Our CompositionRef records the governance state at decision time — the delegation chain, revocation status, and action boundaries that were in effect.

Receipt type	Question	Example
Execution receipt (HELM)	"Was this action allowed?"	verdict: "allow" for github.create_issue
Governance anchor (ours)	"Was the agent's governance state healthy when it was allowed?"	composition_ref: sha256:... proving delegation was intact

Integration proposal

Extend external_decision_receipt.v1 with an optional governance anchor field:

{
  "schema_version": "external_decision_receipt.v1",
  "kind": "external_decision_receipt",
  "format_id": "helm_external.v1",
  "receipt_id": "edr-governance-0001",
  "action": "github.create_issue",
  "verdict": "allow",
  "subject": "did:helm:agent:demo",
  "decision_time": "2026-06-19T12:00:00Z",
  "signature_algorithm": "Ed25519",
  "signature": "...",
  "signing_key_id": "governance-key-1",
  "source_vendor": "agent-os",
  "governance_ref": "sha256:a91887274119578c088014da82119ff39eb27da460733373e9c21726e989958b",
  "governance_model": "composition_ref.v1",
  "quasar_trust": 0.85
}

Key properties:

• governance_ref — SHA-256 hash of the composition state (action_ref + delegation_ref + revocation_ref)
• governance_model — identifies the composition protocol version
• quasar_trust — optional trust score (0-1) from verifier policy layer, independently verifiable

Through HELM's existing helm-ai-kernel import receipt mechanism, this can produce an EvidencePack containing both the execution decision and governance anchor.

Questions

1. Would adding optional governance_ref to external_decision_receipt.v1 be acceptable, or would a separate receipt kind be preferred?
2. Interested in a helm-ai-kernel import governance adapter?
3. Constraints on custom fields in the normalized schema?

What we can contribute

• Reference adapter (Python): CompositionRef JSON → helm_external.v1 format
• Governance state verification CLI
• 28/28 test vectors for the profile constraint layer