Cross-validate against the draft-farley-acta-signed-receipts conformance vectors?

[tomjwxf]

Hi, author of ai-agents-for-beginners Lesson 18 (and the receipt format it teaches) here. Saw the production reference you added to the lesson and the cross-validation work in this repo (the APS round-trip and CTEF vectors).

An invitation along the same lines: the receipt format the lesson follows, draft-farley-acta-signed-receipts (rev 02 on the datatracker), has a shared conformance suite at ScopeBlind/agent-governance-testvectors: 14+ implementations cross-verify byte-identical canonical output, including restraint-receipt vectors (denied and deferred actions produce verifiable receipts too) and negative cases.

If nobulex's Ed25519 + JCS envelope round-trips those vectors, open a PR to the registry and I will list nobulex in the draft's implementation-status section. That gives your receipt layer standing in the IETF process, independent of the Trust Capital layer you build above it, and it makes the Lesson 18 cross-reference concrete: readers could verify a nobulex receipt and a protect-mcp receipt with the same tooling.

Happy to help map fields if anything in the envelope does not line up 1:1.

[github-actions]

🔴 Contributor Check: HIGH

Check	Result
Profile	HIGH
Credential	LOW
Cluster	UNKNOWN
Overall	HIGH

Automated check by AGT Contributor Check.

[arian-gogani]

Thanks for reaching out, and for the invitation. IETF implementation-status listing alongside the OWASP AST09 normative inclusion would be a strong combination.

I will run the ScopeBlind/agent-governance-testvectors suite against Nobulex's Ed25519 + JCS envelope this week. The canonicalizer is byte-identical to the Python rfc8785 reference (conformance vectors at tests/conformance/jcs-rfc8785-cross-impl.test.ts), so the round-trip should hold if the envelope shape aligns.

Two quick questions before I start:

1. Does the conformance suite expect a specific envelope wrapper around the JCS-canonical payload, or is it testing the canonical output + signature independently?

2. For the restraint-receipt vectors (denied/deferred actions): Nobulex already produces DENY receipts with full preimage and signature. Is the conformance shape checking for a specific terminal_state field, or is the denied-before-dispatch property tested structurally (absence of outcome receipt for a given attempt_id)?

Will open the PR to the registry once the vectors pass.