Hi Ben,

Thanks for the careful read, especially on the ParamKey allocation question.

On (1), you are right. ParamKeys are service-connection hints, not indirection 
pointers. Using them to reference a document at a different URL is the wrong 
shape for SVCB. I will drop the atn and atn-digest SVCB ParamKey registrations 
from -01.

On (2), this is convergent with feedback I received from Aijun Wang on the dmsc 
thread, and it matches my own reassessment of -00. -01 will reposition ATN 
around HTTP-resource identity using a .well-known origin discovery pattern:

• Bootstrap is per-origin, not per-agent. The client fetches 
https://{origin}/.well-known/atn (RFC 8615) to get an Index Document.
• The Index Document is signed with JWS, lists the agents available at the 
origin, and carries origin-level capability declarations.
• Each agent is an HTTP resource at its own URL under the origin (e.g., 
https://example.com/agents/alice), returning the Capability Manifest and the 
other per-agent artifacts.
• Many agents share one origin. Creating, modifying, or removing agents 
requires no DNS changes.

DNS becomes one optional binding for the cross-organizational, domain-keyed 
case. The TXT record collapses to a thin announcement: “ATN is supported at 
this origin, see the well-known URI.” It no longer points at individual agents 
and no longer carries integrity hints. That is the part of -00 that earned your 
criticism, and rightly so.

On (3), the new architecture sharpens the JWS justification. The Index Document 
is exactly the artifact that needs offline-verifiable provenance. It gets 
cached, mirrored, fed into SCITT transparency services, and reconstructed 
during audit long after the original TLS connection has closed. HTTPS 
authenticates the fetch; JWS authenticates the content for everything 
downstream. -01 will make JWS on the Index Document MUST, not RECOMMENDED, 
regardless of DNSSEC posture.

To your underlying point on whether DNS is the right discovery substrate at 
all: -01 reframes ATN as discovery-agnostic, with the four artifacts and the 
handshake as the normative core. .well-known on origin is the default binding 
for HTTP-native agents. DNS is one optional binding among several.

I will post -01 before IETF 126 with these changes. Happy to discuss further 
before then or at the dnsop session.

Thanks,
Enrique

On May 18, 2026, at 11:29 AM, Ben Schwartz <[email protected]> wrote:


There have been several messages asking similar questions on DNSOP recently.  
Maybe we should write an FAQ for SVCB...

As I wrote in previous discussions on this topic [1][2]:

1. SVCB generally assumes that you are connecting to a service identified by 
(part of) a URI, including a scheme.  What is the URI in this case?  If there 
is none, SVCB is likely not the right solution.
2. It sounds like you are assuming that agents are identified by domain names, 
but are operating in a HTTP-centric world.  In an HTTP-centric world, it's 
probably better to let agents be identified by an HTTP resource, so that they 
can be created and modified more easily.  (There are also privacy advantages.)
3. It looks like this proposal is trying to fetch HTTP resources that are 
authenticated by a non-HTTP signature chain.  This is a lot more complicated 
than a solution using standard HTTP server authentication, and the benefit is 
not clear to me.

--Ben Schwartz


[1] 
https://mailarchive.ietf.org/arch/msg/dnsop/8v6J897lNyVil4zW3hsR8xGbExw/<https://urldefense.proofpoint.com/v2/url?u=https-3A__mailarchive.ietf.org_arch_msg_dnsop_8v6J897lNyVil4zW3hsR8xGbExw_&d=DwMFaQ&c=euGZstcaTDllvimEN8b7jXrwqOf-v5A_CdpgnVfiiMM&r=XXqibAFiYDdfWNiaTQ7_UQfyRgIxsoUr07awOSGSXHs&m=YlADFbbUDAGEy3h56uLwmMfK4kjj496DNWUDM0KKLVQLVTuYrgOa5AxzPSx43Ees&s=FaHZjfkdSrMuyambkRbvKx0AHYPnKyd-Djlyy1FNB9s&e=>
[2] 
https://mailarchive.ietf.org/arch/msg/dnsop/EUpxbLgilznpyJgaaR7GZwleqOw/<https://urldefense.proofpoint.com/v2/url?u=https-3A__mailarchive.ietf.org_arch_msg_dnsop_EUpxbLgilznpyJgaaR7GZwleqOw_&d=DwMFaQ&c=euGZstcaTDllvimEN8b7jXrwqOf-v5A_CdpgnVfiiMM&r=XXqibAFiYDdfWNiaTQ7_UQfyRgIxsoUr07awOSGSXHs&m=YlADFbbUDAGEy3h56uLwmMfK4kjj496DNWUDM0KKLVQLVTuYrgOa5AxzPSx43Ees&s=2RpQ8xA8jXzfF6y9LoW6cHGDxPZiUOqI1lAdLaMVBBs&e=>
On Sun, May 17, 2026 at 6:15 PM Enrique Somoza 
<[email protected]<mailto:[email protected]>> wrote:
Hi dnsop, This is a cross-post pointer from my dmsc thread earlier today on 
draft-somoza-atn-agent-trust-negotiation-00. The draft sits primarily in dmsc 
but registers SVCB ParamKeys, so I wanted to give dnsop visibility on the 
DNS-layer content
Hi dnsop,

This is a cross-post pointer from my dmsc thread earlier today on 
draft-somoza-atn-agent-trust-negotiation-00. The draft sits primarily in dmsc 
but registers SVCB ParamKeys, so I wanted to give dnsop visibility on the 
DNS-layer content specifically.

The dnsop-relevant part is small. The draft requests two SVCB ParamKey 
registrations:

  atn         URI         Absolute HTTPS URI pointing to an ATN Index Document
  atn-digest  hex-string  SHA-256 digest of the Index Document for integrity 
binding

It also proposes the same two names as TXT key=value tags for compatibility 
with existing TXT-based agent discovery records (notably AID).

The integrity model: the digest is bound to the content at the URL, not the DNS 
lookup itself. This is the same model as Subresource Integrity in web 
platforms. DNSSEC remains useful for authenticating the binding; the digest 
covers the artifact after fetch. The draft requires JWS signature on the Index 
Document when DNSSEC is not validated, and makes it RECOMMENDED but not 
REQUIRED when DNSSEC is in use.

Questions where I would value dnsop input:

1. The name pattern. "atn" and "atn-digest" mirrors patterns in existing 
TXT-based discovery drafts. Is there a dnsop preference for the integrity field 
name (e.g., "atn-i", "atn-sha256", etc.) before this goes anywhere near IANA?

2. Allocation path. The draft is Experimental. Is the right path: (a) 
provisional allocation from the private-use range during the Experimental 
period, (b) early allocation with WG sponsorship, or (c) wait until 
standards-track maturity?

3. SVCB vs TXT. The draft offers both forms for compatibility with the 
discovery proposals already in flight (AID uses TXT; DNS-AID uses SVCB). Should 
ATN deprecate the TXT form and require SVCB given dnsop's general direction?

4. DNSSEC interaction. The current language ("RECOMMENDED but not REQUIRED when 
DNSSEC is in use") may be too permissive. Should JWS-on-Index be MUST 
regardless?

The draft itself is at:
  
https://datatracker.ietf.org/doc/draft-somoza-atn-agent-trust-negotiation/<https://urldefense.proofpoint.com/v2/url?u=https-3A__urldefense.com_v3_-5F-5Fhttps-3A__datatracker.ietf.org_doc_draft-2Dsomoza-2Datn-2Dagent-2Dtrust-2Dnegotiation_-5F-5F-3B-21-21Bt8RZUm9aw-215LBAuicD5Zum5Csa-2DGkjTXR-5FTGt1AazU8V9MEbNqSPz-2DybsAYd-5FhxPXryqdXN-2DZ5LCv-5FoVJR-5F9tjMA-24&d=DwMFaQ&c=euGZstcaTDllvimEN8b7jXrwqOf-v5A_CdpgnVfiiMM&r=XXqibAFiYDdfWNiaTQ7_UQfyRgIxsoUr07awOSGSXHs&m=YlADFbbUDAGEy3h56uLwmMfK4kjj496DNWUDM0KKLVQLVTuYrgOa5AxzPSx43Ees&s=GYpToxflICiU3LGaMMHprGjiaBz44FaMv0WerdXbA8k&e=>

The DNS Binding section is Section 4 and is short.

Thanks,
Enrique
_______________________________________________
DNSOP mailing list -- [email protected]<mailto:[email protected]>
To unsubscribe send an email to 
[email protected]<mailto:[email protected]>
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to