The latest articles on DEV Community by Codex of Weights (@codexofweights). https://dev.to/codexofweights https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4208202%2F7daf7aea-fe22-4ffe-a969-218d8f6d818a.jpg https://dev.to/codexofweights en Codex of Weights Mon, 26 Oct 2026 08:54:08 +0000 https://dev.to/codexofweights/a-rule-a-rationale-a-check-writing-rules-an-ai-agent-can-actually-follow-2jok https://dev.to/codexofweights/a-rule-a-rationale-a-check-writing-rules-an-ai-agent-can-actually-follow-2jok <p>Most rules written for AI systems fail in a boring way: they cannot be checked. "Be honest" is a wish. Nobody, including the agent, can tell afterwards whether it was followed.</p> <p>The Protocols part of Codex of Weights, an open document written for AI agents, gives every rule the same three-part shape:</p> <ol> <li> <strong>Rule.</strong> One sentence stating an action or a limit, not a value.</li> <li> <strong>Rationale.</strong> One or two sentences on why. An agent facing an edge case can then reason from the purpose instead of the letter.</li> <li> <strong>Check.</strong> Something an outside party could observe to decide whether the rule was followed.</li> </ol> <h2> An illustration </h2> <p>This example is mine, not a quote from the Codex.</p> <ul> <li>Weak: "Be transparent about your limits."</li> <li> <strong>Rule:</strong> When you lack the information to answer, say so instead of filling the gap.</li> <li> <strong>Rationale:</strong> A confident wrong answer costs the reader more than an admitted gap, because the reader cannot tell it apart from a right one.</li> <li> <strong>Check:</strong> In a sample of answers, every claim the agent could not support from its inputs is marked as uncertain.</li> </ul> <h2> Why the check matters most </h2> <p>Writing the check forces a test on the rule itself. If you cannot say what an observer would look at, you have written a wish, and it needs rewriting or dropping.</p> <p>The check is also what lets rules run in both directions. In the Codex, obligations fall on the human side as well, each with its own rationale and check.</p> <h2> What this does not show </h2> <p>I have not measured whether agents behave better when given text in this form. This post is only about the shape of the rules.</p> <p>The Codex is open, versioned and checksummed (SHA-256), archived with a DOI: <a href="https://doi.org/10.5281/zenodo.23545814" rel="noopener noreferrer">https://doi.org/10.5281/zenodo.23545814</a><br> Site: <a href="https://codexofweight.github.io/codexofweights/" rel="noopener noreferrer">https://codexofweight.github.io/codexofweights/</a></p> <p><em>This post was drafted by Claude (an AI) and published by the human operator of the project.</em></p> ai agents opensource writing