Governance in One File

Rules before tokens

Most AI governance happens after the fact: someone reviews logs, finds the agent did something it shouldn't have, and writes a postmortem. Governance in Cromus is the opposite — the rules are declared, validated, and scored before any token is spent.

The declaration lives in one file: ETHOS.md. It states what an agent may do, what is prohibited, what requires confirmation, and how violations escalate. Because it's a file — not a policy PDF — it can be validated deterministically, versioned in git, and enforced by any compliant runtime.

What the validator checks

The ETHOS validator scores three things:

• Completeness — are the required sections present (identity, boundaries, escalation)?
• Compliance — do declared behaviors match a coherent behavioral profile?
• Governance — are the risk controls (confirmation gates, prohibited actions) actually specified?

It also flags gaps as errors and warnings, and estimates the ethics-related Croms your workflow carries while those gaps stay open.

Do It: break it, then fix it

The panel below loads the default Cromus ETHOS baseline and validates it. Then make one edit — the suggested one-liner, or your own — and re-validate. Watch the flag count and score change. That feedback loop is the whole point: governance you can edit and re-score in seconds gets maintained; governance in a PDF does not.