Is there a global standard for independently verifying AI agents, tool calls, token usage, costs, and task correctness?

Hi everyone!

I'm building Korako Yolawani, a human-first AI assistant that coordinates AI agents and tools while keeping the user in control.

I'm researching whether an internationally recognized, vendor-neutral standard already exists to evaluate the complete execution of an AI agent's task.

Specifically, I'm looking for a framework that can:

1. Verify that an agent actually completed the task correctly, with independent evidence.
2. Track every model call, tool call, token, retry, and associated cost.
3. Evaluate execution efficiency without sacrificing quality or reliability.
4. Prevent unauthorized actions, unexpected expenses, data leaks, and harm to users.
5. Enforce hard spending limits, permissions, STOP controls, and human approval.
6. Produce independently auditable records across different AI providers and agent frameworks.

I'm currently exploring IEEE P3777, ISO/IEC AI standards, NIST, OWASP, and OpenTelemetry.

Our guiding principle is:

PREPARED ≠ EXECUTED ≠ VERIFIED.

My question to the community:

Is there already one practical open-source framework or industry standard that combines all these capabilities? If not, which existing standards and tools would you recommend integrating?

I'm especially interested in real-world implementations, reproducible evaluations, and free or open-source solutions.

Thank you!

Miha Tavčar
Founder, Korako Yolawani