Small MCP-first post-work receipt reference for agent runs

[0xddneto]

A small post-work receipt reference for Semantic Kernel runs

I am building AIPOU as an MCP-first protocol for people working with AI agents. It creates a private signed receipt after a meaningful work unit completes and exposes the opaque ID as workReceiptId.

The proposed Semantic Kernel surface is intentionally small:

1. start a receipt at an agent/process run boundary;
2. at completion, retain only local hashes plus provider/model metadata;
3. attach the returned workReceiptId as optional run, process, trace, or delivery metadata.

The host keeps its own authority, workflow, tool, and audit semantics. AIPOU is external issuer_asserted evidence: it does not prove hidden AI use, task quality, payment, or provider endorsement. It should not enter a host's native chain of custody as a first-party record.

The first runnable demo uses Local Receipt Mode and synthetic data only. It needs no wallet, funds, token transfer, claim, customer data, raw prompt upload, or network access:

• Demo: https://github.com/0xddneto/AI-Proof-of-Us/tree/main/examples/local-receipt-mode
• Lifecycle adapter: https://github.com/0xddneto/AI-Proof-of-Us/blob/main/docs/framework-lifecycle-adapter.md
• Boundaries: https://github.com/0xddneto/AI-Proof-of-Us/blob/main/docs/work-receipt-boundaries.md

I would value feedback on the cleanest Semantic Kernel extension point: process/run metadata, OpenTelemetry activity attributes, or a result/audit export. Optional validator-approved rewards for the humans using agents are a separate AIPOU path and are deliberately out of scope for this integration.