We have plenty of guardrails for AI prompts, we have almost no standard for proving what an agent actually did after execution.
RIO is a 'Layer 3' protocol that binds the Intent Hash to the Execution Hash in a tamper-evident ledger.
Zero Dependencies: The Node.js and Python reference impls use only standard libraries (node:crypto/hashlib).
Tamper-Evident: It uses a SHA-256 hash chain where any modification, deletion, or reordering breaks the linkage.
Extensible: It supports a standard 3-hash chain (Intent → Execution → Receipt) but can expand to a 5-hash chain if you need to prove human-in-the-loop approval.
I’ve included a CLI verifier and 29 conformance tests in the repo. Curious to hear what people think about the schema or the ledger format."
We have plenty of guardrails for AI prompts, we have almost no standard for proving what an agent actually did after execution. RIO is a 'Layer 3' protocol that binds the Intent Hash to the Execution Hash in a tamper-evident ledger. Zero Dependencies: The Node.js and Python reference impls use only standard libraries (node:crypto/hashlib). Tamper-Evident: It uses a SHA-256 hash chain where any modification, deletion, or reordering breaks the linkage. Extensible: It supports a standard 3-hash chain (Intent → Execution → Receipt) but can expand to a 5-hash chain if you need to prove human-in-the-loop approval. I’ve included a CLI verifier and 29 conformance tests in the repo. Curious to hear what people think about the schema or the ledger format."