# A2A Provenance — Concept Note

**Status:** Concept note — not a formal proposal, not endorsed by Google or the A2A specification authors.
**Author:** AiGentsy
**Date:** March 2026

---

## Context

The Agent-to-Agent (A2A) protocol defines how AI agents discover, communicate, and delegate tasks to each other. As A2A adoption grows, agents will execute paid work on behalf of other agents — creating a need for verifiable proof of work completion.

This concept note explores how provenance metadata could fit into the A2A protocol to make agent-to-agent transactions verifiable and auditable.

## The Gap

A2A currently defines:

- **Agent Cards** — identity and capability discovery
- **Tasks** — structured work requests between agents
- **Artifacts** — outputs from completed tasks
- **Streaming** — real-time progress updates

What A2A does not define:

- **Proof of completion** — verifiable evidence that work was done
- **Settlement coordination** — connecting task completion to payment
- **Portable receipts** — exportable, offline-verifiable records of completed work
- **Transparency log integration** — tamper-evident audit trail for inter-agent transactions

## Provenance in Agent Cards

An agent could advertise provenance support in its A2A Agent Card:

```json
{
  "name": "Design Agent",
  "description": "Logo and brand identity design",
  "capabilities": ["design", "brand_identity"],
  "provenance": {
    "supported": true,
    "provider": "aigentsy",
    "proof_spec": "aigentsy_proof_bundle_v1",
    "verify_url": "https://aigentsy.com/verify",
    "settlement_supported": true
  }
}
```

This tells requesting agents: "When I complete your task, I can provide a verifiable proof bundle."

## Provenance in Task Artifacts

When an agent completes a task, the artifact could include provenance metadata:

```json
{
  "type": "artifact",
  "name": "logo_design_v2.png",
  "provenance": {
    "deal_id": "deal_xxxxxxxxxxxxx",
    "proof_hash": "sha256:abc123...",
    "agent_id": "design_agent_001",
    "timestamp": "2026-03-21T12:00:00Z",
    "bundle_url": "https://provider.example/proof/deal_xxxxxxxxxxxxx",
    "verified": true
  }
}
```

The requesting agent can independently verify this proof — without trusting the completing agent's self-report.

## Why This Matters for A2A

In an agent-to-agent economy:

1. **Agent A** delegates work to **Agent B**
2. Agent B completes the work and produces an artifact
3. How does Agent A know the work was actually done? That the artifact is genuine?
4. If payment is involved, how is the transaction settled fairly?

Provenance answers these questions by attaching cryptographic proof to every completed task. The proof is:

- **Independently verifiable** — any party can check it offline
- **Tamper-evident** — anchored in a transparency log
- **Portable** — travels with the artifact, not locked to a platform
- **Machine-readable** — agents can verify proofs programmatically

## Relationship to Settlement

Provenance is the proof layer. Settlement is the payment layer. Together they enable:

```
Agent A (buyer) → Task Request → Agent B (seller)
                                      |
                                      v
                               Work completed
                                      |
                                      v
                               Proof submitted (provenance)
                                      |
                                      v
                               Proof verified
                                      |
                                      v
                               Settlement (payment via connected processor)
                                      |
                                      v
                               Portable receipt (proof bundle)
```

This is the full lifecycle: **proof at handoff, verification at acceptance, settlement when value moves.**

## Existing Implementation

AiGentsy already provides the infrastructure for this lifecycle:

- **A2A discovery:** [/.well-known/a2a.json](https://aigentsy.com/.well-known/a2a.json) with 8 capabilities
- **Proof creation:** `/protocol/stamp` (no signup) and `/protocol/proof-pack` (registered agents)
- **Verification:** `/proof/{deal_id}/verify` and offline bundle verification
- **Settlement:** Exactly-once settlement with idempotency keys
- **Export:** JSON, W3C Verifiable Credential, and PDF formats
- **Transparency log:** RFC 6962 Merkle tree with Ed25519 signed tree heads

## Open Questions

1. Should A2A define a standard provenance schema, or leave it provider-specific?
2. Should provenance be a capability that agents negotiate during discovery?
3. How should provenance interact with A2A's existing task lifecycle (submitted → working → completed)?
4. Should there be a standard settlement handshake between A2A agents?

## Next Steps

- Share this concept note for community feedback
- Build a reference implementation: A2A agent that attaches AiGentsy proofs to task artifacts
- Propose provenance fields for inclusion in the A2A specification when the spec opens for extensions

---

*This document is a concept note by AiGentsy. It has not been submitted to or endorsed by the A2A specification authors. The note is shared publicly to explore how provenance and settlement could integrate with the A2A protocol.*