Academic paper proposing topology-independent agent identity with A2A integration. feedback welcome

[rrrodzilla]

Hi,
I recently published a paper on arXiv proposing a URI scheme for decentralized agent identity in multi-agent systems: https://arxiv.org/abs/2601.14567

The work addresses a gap I encountered when building agent systems: agents lack stable identities independent of their deployment topology. When an agent moves hosts, changes providers, or scales across instances, URL-based identity breaks or requires central coordination to maintain.

Proposed solution

A URI scheme that provides:

•	Topology-independent identity — agents have stable identifiers regardless of where they run
•	Capability-based discovery — find agents by what they can do, not where they are
•	Decentralized resolution — DHT-based lookup without central registries
•	Cryptographic attestation — trust roots sign capability claims via PASETO tokens

The URI structure:
agent://trust-root/capability-path/agent-typeid

For example:

agent://anthropic.com/assistant/chat/llm_chat_01h455vb4pex5vsknk084sn02q agent://acme.corp/workflow/approval/rule_fsm_01h5fskfsk4fpeqwnsyz5hj55t

Relationship to A2A

The paper includes a section on A2A integration, proposing:

1.	Agent Card extensions — adding a uri field to Agent Cards that provides a stable identifier decoupled from the service endpoint URL
2.	Attestation via PASETO — addressing the roadmap item around formalizing authorization schemes and credentials
3.	Capability-based discovery — complementing the existing well-known endpoint approach with DHT-based lookup for scenarios where you want to find agents by capability rather than already knowing their endpoint
4.	Backward compatibility — non-URI-aware agents continue to work; the URI is additive

The core insight is separating identity (the URI, which is stable) from location (the endpoint URL, which can change). An agent’s URI remains constant even as it migrates between providers or scales across regions.

Why I’m posting here

I noticed the roadmap mentions:
“Formalize inclusion of authorization schemes and optional credentials directly within the AgentCard”

The PASETO attestation model in the paper is one approach to this. Trust roots sign tokens binding an agent’s TypeID to its claimed capabilities, which can be verified without contacting the trust root on every request.

I’d welcome feedback on:

1.	Whether this identity model aligns with A2A’s design philosophy
2.	Potential conflicts or redundancies with planned A2A features
3.	Interest in the capability-based discovery approach as a complement to well-known endpoints
4.	Any concerns about the Agent Card extension proposal

For context, I maintain rusty_paseto and mti (Rust implementations of PASETO and TypeID respectively), so I’ve spent time with the underlying specs; this paper grew out of thinking about agent identity problems where those primitives were a natural fit.

Happy to discuss any aspect of the design, and thanks for building A2A—it’s addressing a real need in the ecosystem.