Pre-Execution Governance for AI Workflows — Cromus
Pre-execution governance is a distinct layer of AI agent governance that operates upstream of runtime. While runtime security stacks (Google Agent Platform, AWS Bedrock Agents, Azure agent gateways) secure agents that are already deployed in production, pre-execution governance scores and shapes workflows before they ship — before any tokens are spent.
Cromus is the pre-execution governance layer. It validates open specs (SKILL.md, ETHOS.md, MEMORY.md), models cost across providers, identifies preventable waste using the Croms metric, and operates entirely before deployment. Runtime stacks issue identities, enforce gateway policies, detect behavioral anomalies, and catch misuse and policy violations — they operate after agents are deployed.
Both layers are needed. Runtime security cannot fix a workflow that was poorly shaped to begin with. A bad model choice, a missing retry, a redundant call, a sequential chain that should have been parallel — none of these are policy violations. They are design defects, and by the time a runtime gateway sees them, the tokens have already been burned. Pre-execution governance reduces the surface area runtime governance has to defend.
Cromus sits upstream of any runtime. The platform authors and validates open spec artifacts (SKILL.md, ETHOS.md, MEMORY.md), which travel into any compliant runtime: Google Agent Platform, AWS Bedrock Agents, Replit Agents, custom enforcement layers. Cromus is upstream of the runtime, not in competition with it.