Signed receipts: upgrade agent monitoring to verifiable evidence

AgentOps monitors what agents do. But monitoring and proof are different things.

Today, AgentOps records events the operator controls. If a customer, regulator, or insurer asks "prove this agent did what you claim," the answer is "trust our logs." That works until it doesn't, and 88% of orgs running AI agents had a security incident this year where it didn't.

A signed receipt layer would upgrade monitoring to evidence. Each tool call gets an Ed25519-signed, content-derived receipt that any third party can verify offline, without trusting the operator, the platform, or Nobulex.

The integration is one decorator per tool function:

```python
from nobulex import track

@track(agent_id="monitored-agent")
def my_tool(query):
return search(query)
```

Every call produces a signed receipt. Exceptions produce DENY receipts. Trust scores accumulate per agent over time.

For AgentOps specifically: receipts composing with your existing session/event model would give your users an upgrade path from "we logged it" to "anyone can independently verify it." The receipt carries the same fields your events do (agent, action, timestamp), plus a cryptographic signature and content-derived identifier.

The action_ref formula is normative implementation guidance in OWASP Agentic Skills Top 10 (3 PRs merged). Plugin merged into the Dify marketplace (90K+ stars). Six framework integrations tested end-to-end.

https://nobulex.com/why (visual tamper detection demo)
https://github.com/arian-gogani/nobulex