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."