Hello OAuth Working Group, I am developing a small research prototype called Valora Proof and would appreciate technical feedback from participants working on OAuth / AI-agent authorization.
I am not proposing a new OAuth standard, and I am specifically trying to avoid re-inventing existing authorization mechanisms. The v0.1 prototype explores a narrower end-to-end execution/evidence problem: - a human issues a narrowly scoped, one-use authorization; - the AI agent does not hold the external execution credential; - a separate policy/gateway layer enforces the authorization; - revocation vs execution-start ordering is explicit; - retries, response loss, restart, and RESULT_UNKNOWN are modeled; - external execution evidence is signed; - evidence is exported in a portable bundle; - a separate verifier evaluates claims using an independently supplied Trust Config. We already treat fine-grained authorization, PDP/PEP separation, OAuth-style delegated authority, and agent identity as areas where existing standards should be reused rather than claimed as Valora-specific novelty. I would value a focused review on: 1. which authorization semantics should be replaced by or aligned with OAuth/AuthZEN/WIMSE work; 2. whether the one-use authorization / execution-start boundary is clear; 3. whether our Trust Config assumptions are too proprietary or weak; 4. whether the evidence/recovery layer addresses a real gap, or duplicates existing work; 5. the smallest blocker you would require us to fix before a low-risk pilot. The prototype currently uses a mock merchant only: no real money, no production credentials, no blockchain requirement. The repository is private, so I can send a concise review packet and selected specification/source material directly to any volunteer reviewer. Critical feedback is preferred over endorsement. Thank you, Toshio Nakanishi Valora AI / Valora Proof
_______________________________________________ OAuth mailing list -- [email protected] To unsubscribe send an email to [email protected]
