Replit Agent Skills & Governance — Cromus
Cromus is the pre-execution governance layer around Replit Agent. Replit is where the work runs; Cromus helps teams choose a sensible model and effort, validate a portable skill, and understand what is governed before the first token is spent. Cromus does not replace Replit Agent or change a user's Replit settings.
The Replit Agent Mode Advisor evaluates scope, reasoning, risk, verification, and reversibility, then recommends Free, Power, or Max plus low, medium, or high effort guidance. It cannot read the user's account plan or usage, override Replit platform routing, or guarantee a particular model. The user must choose the setting in Replit's Agent settings or Workspace Settings.
Cromus tracks Replit's current model selector as a two-axis choice: 18 documented model rows across Lite, Economy, and Power, plus five effort levels — low, medium, high, extra high, and max. Credit rates are explicitly marked unverified where Replit has not published them; Cromus never fabricates a dollar estimate. Skill Studio can analyze and validate a workflow for the Replit Agent target, prepare Replit-compatible configuration output, and preserve the governed SKILL.md source. Legacy mode, Turbo, and high_effort fields are migrated with warnings rather than silently rewritten.
The broader governance workflow is portable. SKILL.md defines capability, ETHOS.md defines behavioral rules, MEMORY.md defines portable context, and AGENTS.md gives coding agents project context. The same artifacts can be checked by the free validators, the Specialist in CI, Cromus Gate, or deterministic MCP tools. Cromus can recommend, validate, score, compile, and flag drift; Replit controls the final Agent setting, account model availability, platform routing, workspace permissions, billing, and execution.