--- title: Why Cromus — Governance That Travels With Your Workflows description: Cromus governs and scores AI workflows before tokens are spent — over portable open specs (SKILL.md, ETHOS.md, MEMORY.md) that travel into any runtime. See the unique features that make it a complete workflow intelligence tool. canonical: https://cromus.ai/why-cromus source_html: https://cromus.ai/why-cromus --- # Why Cromus > Governance that travels with your workflows. Cromus governs and scores AI workflows **before tokens are spent** — over open specs that travel into any runtime. Most tools force a trade-off: lock down governance, or stay portable. Cromus refuses it. The governance and the portability are the same artifact. --- ## Why governance and portability stop fighting here Runtime governance locks rules to a platform — strong control, but your workflow can no longer leave. Portable artifacts move freely — but usually carry no governance with them. Cromus collapses the trade-off by putting the governance **inside** the portable artifact. Cromus authors and validates the open specs — **SKILL.md** for capability, **ETHOS.md** for behavior, **MEMORY.md** for portable memory — and scores them before execution. The budget envelopes, model-governance locks, and behavioral rules ride along in the spec. When the spec lands in Anthropic, Google, AWS Bedrock, Replit, or a custom runtime, the governance is already there. Cromus never sits between the agent and the model; `scope: "enforce"` is rejected by design. --- ## What's improved lately — stronger governance, still portable The throughline of recent work is pre-execution governance over portable specs. Each addition tightens control without trapping your workflow on a single platform. - **Champion / Challenger Experiments** — reality-gated A/B testing of skill versions. Zero LLM calls, no auto-arm-assignment, no auto-promotion, no p-values, and Croms is never an objective. A human reads an observed verdict and decides — the governed unit stays a portable workflow version, never a locked runtime config. - **Pre-Execution Budget & Platform Governance (Layer 1/2/3)** — author-declared cost/token envelopes and credit-pool platform context live inside SKILL.md and ETHOS.md. `scope: "enforce"` is rejected — it's pre-flight only, so the governance travels in the open spec into any compliant runtime. - **Single source-of-truth scoring** — the same validation engine powers the public validator, the MCP tools, and the in-app readiness tile. A governance score means the same thing everywhere, and ETHOS findings never leak into the Croms number. - **Hosted MCP server — zero LLM, no retention** — six deterministic tools usable from Claude Cowork, Claude Code, Codex, or any MCP client. Every response carries an attribution block under a no-retention contract, and the transport is stateless. - **Skill Studio platform selector** — compile one governed SKILL.md to six targets (Cromus Native, Replit Agent, AgentSkills.io, OpenClaw, Claude Managed Agents, Google AI Studio), stripping and normalizing per-platform fields. One governed artifact, many destinations. --- ## The complete workflow intelligence toolkit Everything below exists to govern and measure portable artifacts before they execute. ### Measurement & scoring - **Croms™** — quantifies preventable AI workflow waste with a transparent scoring matrix. - **4-Step Framework** — Classification → Baseline Cost → Croms → Total Cost of Workflow Ownership (TCWO). - **Cost Simulator** — priced against a verified, regularly updated model registry. - **Workflow Health Score** and **AI Compatibility Engine** — SOPs scored for AI readiness. ### Closed-loop learning - **Reality Check Loop** — Predict → Execute → Log → Learn, with Adoption & Confidence engines. - **Champion / Challenger Experiments** — reality-gated A/B of skill versions. - **CSV Usage Uploader** — real provider usage in, cost recomputed from registry prices (never trusted). ### Authoring & governance - **SOP Compiler** — turns Standard Operating Procedures into portable SKILL.md files. - **Open Spec Stack** — ETHOS.md, SKILL.md, MEMORY.md with three free validators. - **Pre-Execution Budget & Platform Governance** (Layer 1/2/3) plus model-governance locks. - **Skill Studio** — universal analyzer, visual canvas, and a 6-platform selector. ### Determinism & trust - **Mock AI mode** — predictable, reproducible scoring with no external LLM dependency. - **Scoring consistency** across the web validator, MCP tools, and in-app surfaces. ### Interoperability & distribution - **Hosted Cromus MCP server** — zero LLM calls, OAuth 2.1 + PKCE, Cowork & Codex connectors. - **Open Workflow Ecosystem** — export to LangGraph, CrewAI, AutoGen, n8n, and more. - **Template libraries** — skill templates, workflow templates, sample packs & startup packs. ### Operations & go-to-market - **Cromus Inbox** (proactive recommendations), **ROI Dashboard**, and notification workflows. - **Croms™ Diagnostic Quiz** and the **Cromus Guide** in-product assistant. - **White Label** agency workspaces and upload security scanning. --- ## Where Cromus fits in your stack Cromus is the upstream governance and scoring layer. It authors and validates the open specs; the specs travel into any compliant runtime; the runtime enforces them at execution. **Cromus (govern & score)** → **Open Specs (SKILL · ETHOS · MEMORY)** → **Any Runtime (Anthropic · Google · AWS · Replit)** Cromus is upstream of the runtime, not in competition with it. --- ## Related concepts - [Pre-Execution Governance](/pre-execution-governance.md) - [Workflow Intelligence as a Service](/workflow-intelligence-as-a-service.md) - [Croms™ — preventable AI workflow waste](/croms.md) - [Open Workflow Ecosystem](/open-workflow-ecosystem.md) - [SKILL.md Validator](/validator/skill.md) - [ETHOS.md Validator](/validator/ethos.md) - [MEMORY.md Validator](/validator/memory.md)