The Scope Creep Automated reviews of new AI-related working group submissions

Update review · November 19, 2026

KYAPay Token Exchange -03 Update

draft-skyfire-oauth-kyapay-token-exchange -03View on Datatracker →

Status: Individual draft (not adopted), Standards Track. Revision -03 published 16 November 2026, two weeks after -02. 15 pages, up from 14. Same four authors.

This is a maintenance revision. It changes four things, and all four are reference or example fixes. The protocol is identical to -02. Of the six items in the -02 review, half of one was addressed.

What got fixed

  • The token reference moved to KYAPay Token -03. The example token now carries hid.id, so it is a valid token under the current format (Section 8).
  • apd can no longer be empty in the direct-method client ID. The token draft makes apd REQUIRED, so the "no apd claim" case was removed (Section 4). Only ori can be an empty segment now.
  • The CIMD section reference is correct. It points at Section 5 of Client ID Metadata Document -03.

What didn't

  • There are still no processing rules for the authorization server. Section 6 says the grant profile means the server "implements the processing rules of this specification." There are none. Nothing says to validate the token, check exp or jti, or which error to return.
  • Audience checking still doesn't work. The example still has tdm set to auth101.dev and a resource of https://mcp.acme.example/. No rule ties aud, tsi, or tdm to the authorization server or the resource.
  • The direct method is still a bearer token redeemable for an access token. cnf is still never mentioned.
  • The CIMD client is still not bound to the token. Nothing ties the CIMD URL to apd or aid.
  • Account matching is still unspecified. The use case text still says the server extracts "email addresses." The example gained hid.id and the prose never mentions it.
  • The move from jwt-bearer to token exchange is still unexplained.
  • Payment tokens, scope, and act are untouched.
  • Security Considerations is still one sentence.
  • None of the four removal suggestions were taken. The kyapay: client ID, both methods, kyapay_issuers_supported, and the kyapay token type URI are all still here.
  • The editorial list is untouched. The collapsed list in Section 4, "KYA Token" versus "KYAPay token," undefined "mission," the unencoded | and :// in the form body, and the matching aud and tsi UUIDs all survive.

What's new

A citation to RFC 10070. The CIMD method's private_key_jwt text now says "as updated by [RFC10070]." That RFC requires a client assertion's aud to be the authorization server's issuer identifier, as the sole value. This is the only new normative-sounding content in the revision.

No mechanisms were added. None were removed.

New problems

The validation rules now exist, in a draft this one doesn't cite. Using KYAPay Tokens -03 has recipient validation steps, tdm checks, and cnf proof-of-possession processing. It says an authorization server declining an exchange responds per RFC 8693 Section 2.2.2. It says the resulting access token should be sender-constrained. It names this draft as the way to do the exchange. This draft has no reference to it at all. An implementer reading only the token exchange spec gets none of those rules.

It is one token revision behind again. The normative reference is KYAPay Token -03. Token -04 was published the same day. The stated reason for dropping the empty-apd case is that apd is REQUIRED, and the claim list in Token -03 still says OPTIONAL. Token -04 is the revision that fixed it.

The introduction disagrees with the token draft. Token -04 describes the exchange as extracting "hid.id, or an email address where it requested one." This draft still says "such as email addresses." The two drafts share authors and a publication date.

The RFC 10070 citation tightens the client's audience and leaves the token's alone. The client assertion must now be audienced to the authorization server. The KYAPay token in the same request still has no audience rule. The stricter check is on the less important JWT.

The direct-method client ID is still different at every target. Token -04 makes aid.id pairwise per Target. The draft cites OpenID Federation and OpenID4VP as precedent for unregistered client IDs. In both of those the ID is stable. Here an authorization server can't use the client ID to recognize the same agent twice across targets, and the draft doesn't say so.

Background of the Authors

No change. Ankit Agarwal, Michael B. Jones, Abhishek Hingnikar, and Jeffrey Hickman are still the authors. Ammar Safdari is still the only listed contributor.

Suggested Feedback

  1. Cite Using KYAPay Tokens normatively and point at its validation steps. The rules have been written. This draft should say which ones an authorization server MUST apply before issuing an access token.
  2. Add the audience rule. State how tdm, tsi, or aud must match the resource parameter or the authorization server. Fix the example so they do match.
  3. Say what the server does with cnf. If the token carries a key, require proof of possession at the token endpoint. If it doesn't, say whether the direct method is allowed at all.
  4. Fix the introduction. Key accounts on iss plus hid.id. The companion drafts already say this.
  5. Repeat the open items from -02. Drop the kyapay: client ID, bind the CIMD client to apd or aid, explain the grant type change, and say which token types are accepted.
  6. Publish the family together. Two revisions in a row have cited a token draft that was superseded the same day.

For related reading, see Using KYAPay Tokens, which now holds the recipient processing rules this draft is missing, and RFC 8693 Token Exchange, which this draft profiles.