Feature: Verifiable action receipts for agent tool calls

Problem

Phidata agents execute tool calls — web search, API calls, database queries, code execution. There's no standard way for the agent to cryptographically prove what it did, what inputs it received, and what it returned.

This matters for:

• Enterprise deployments where compliance requires non-repudiable audit trails
• Agent-to-agent interactions where one agent needs to verify another's work
• Agent commerce (x402 micropayments, Stripe agent checkout) where payment and action must be cryptographically linked

Proposal

Wrap tool execution with Agent Action Receipts (AAR) — Ed25519-signed JSON receipts per action:

# Conceptual integration
class VerifiableAgent(Agent):
    def run_tool(self, tool, args):
        result = super().run_tool(tool, args)
        receipt = sign_receipt(
            agent_id=self.name,
            action={'tool': tool.name, 'args_hash': sha256(args)},
            output_hash=sha256(result),
        )
        self.receipts.append(receipt)
        return result

Each agent run produces a receipt chain — complete verifiable execution history. Any party can verify independently using the public key.

Why now

Mastercard announced Verifiable Intent on March 5, 2026 — the same trust primitive for agent commerce. AAR includes a bidirectional compatibility mapping.

SDK

TypeScript: npm install botindex-aar (live, MIT, single dep) Python: in development

• Spec: https://github.com/Cyberweasel777/agent-action-receipt-spec
• Landing: https://aar.botindex.dev