---
title: ETHOS.md Validator — Free open-spec validator
description: ETHOS.md is an open standard for behavioural governance files that define how AI agents should behave — ethical constraints, fallback behaviours, escalation policies, transparency requirements. The Cromus ETHOS.md validator is free, deterministic, and runs in your browser.
canonical: https://cromus.ai/validator/ethos
source_html: https://cromus.ai/validator/ethos
---

# ETHOS.md Validator

> Free, deterministic validator for the ETHOS.md open specification.

Paste or upload any ETHOS.md file. Get a structured report against the open spec — including behavioural policy, escalation rules, refusal conditions, and transparency requirements. No signup, no API key, no LLM calls.

---

## What ETHOS.md is

ETHOS.md is an open standard for **behavioural governance files**. It defines how an AI agent should behave across the dimensions that matter for safety, accountability, and trust:

- **Ethical constraints** — what the agent must never do
- **Fallback behaviours** — what the agent does when uncertain
- **Escalation policies** — when and how the agent hands off to humans
- **Transparency requirements** — what the agent must disclose
- **Refusal conditions** — when the agent declines a request

ETHOS.md is paired with SKILL.md (capability) and MEMORY.md (retention) as the three open specs that together define a portable, governed AI workflow.

---

## What the validator checks

The Cromus validator runs deterministic checks across:

- **Schema conformance** — required sections, valid policy structure
- **Escalation completeness** — every declared trigger has a defined handler
- **Refusal coverage** — refusal conditions are unambiguous and testable
- **Transparency disclosures** — required disclosures match declared agent capabilities
- **Reference integrity** — referenced SKILL.md skills exist and have matching policies

Every check is deterministic. The same file always produces the same report.

---

## Why behavioural governance needs to be portable

Most runtime safety stacks (Anthropic safety filters, Google Agent Platform guardrails, AWS Bedrock guardrails, Azure content safety) ship their own opinionated policy schemas. If you change runtime, you re-author your governance.

ETHOS.md sits **upstream** of all of them — one portable behavioural policy that compiles down to any runtime's native format.

---

## How to use

- **Web** — paste / upload at [cromus.ai/validator/ethos](https://cromus.ai/validator/ethos)
- **MCP** — call `validate_ethos` on the Cromus MCP server at `https://mcp.cromus.ai/mcp`
- **CI** — same MCP tool, zero LLM calls, no-retention contract, safe for production CI

---

## Related

- [ETHOS.md spec on GitHub](https://github.com/cromus-ai/specs)
- [SKILL.md Validator](/validator/skill.md)
- [MEMORY.md Validator](/validator/memory.md)
- [Pre-Execution Governance](/pre-execution-governance.md)