I have been walking through the governance tools I keep building under API Commons, one a day, and today’s is the one that finally admits something I have been circling for a while: a lot of the APIs being written right now are not being written by people. They are being generated by agents, and the way we govern generated work is still to let the machine produce something and then lint it afterward and tell it everything it got wrong. That is backwards. If I know the rules before I start writing, I write to them. An agent deserves the same courtesy, and that is what Agent Rule Export is for — it takes a governance ruleset and turns it into artifacts an agent can actually read while it authors.
A Spectral ruleset is built to be executed. It is a set of instructions for a linter to run against a finished document, which is exactly the wrong shape for an agent that has not written the document yet. So this tool reshapes the same rules into four things an agent can consume up front. It generates an AGENTS.md governance block — imperative MUST/SHOULD/MAY statements grouped by section for the emerging repo convention that coding agents already look for. It generates a compact system-prompt instruction set to prime an LLM, rules ordered by strength so the strong ones land first. It exports a remediation prompt pack, the per-rule fix guidance keyed by rule id so an agent can look up the exact repair for a violation. And it produces a machine-readable rule digest, a stripped-down checklist an agent can self-check against without carrying any Spectral execution baggage.
The translation that makes this work is quiet but it matters. Severity becomes strength: an error rule becomes a MUST, warn and info become SHOULD, a hint becomes MAY. That is the whole bridge between a linter’s world and an agent’s world, and once you see it you cannot unsee it. Governance was always a set of expectations about how an API should be shaped; we just kept encoding those expectations in a format only a validator could read. Hand the same expectations to the thing doing the authoring and a whole category of after-the-fact correction simply stops happening.
Like the rest of this tooling it runs entirely in your browser, no backend and no account, and anything you paste never leaves the page. You pick a format — OpenAPI, AsyncAPI, Arazzo — optionally narrow by consumer experience or a search term, and copy or download any of the four artifacts. Or you switch the source to your own ruleset, drop in a Spectral rules: map, and export your governance instead of mine. It pairs with the Validator that enforces the rules and the Governance MCP that lets an agent lint on demand: export the guidance here so the agent writes it right the first time, then enforce with the others when you want to be sure.
That is the shift I keep pushing for. Governance should be something you carry into the work, not a gate you hit at the end of it, and that is doubly true now that so much of the work is being done by agents that will follow the rules if you simply give them the rules. This is one more of the API Commons tools I have been writing up this stretch — go try it, point it at a ruleset, and hand the rules to your agent before it starts, not after.
