Bring a governed workflow to Replit Agent.
Replit is where the work runs. Cromus is the layer before it: choose a sensible model and effort, validate the portable skill, and know what is governed before the first token is spent.
The Replit workflow, upstream
One governed artifact. A clearer decision before you build.
Cromus does not replace Replit Agent. It helps you arrive with a workflow that is portable, explainable, and ready for the runtime you choose.
Choose the right starting point
Agent Mode Advisor
The reusable advisor weighs scope, reasoning, risk, verification, and reversibility before recommending Free, Power, or Max, plus a low, medium, or high effort level.
It gives guidance only. You still choose the setting in Replit.
Plan against the current surface
Model Selector Coverage
Cromus tracks the Replit Agent selector as a two-axis choice: 18 documented model rows across Lite, Economy, and Power, with five effort levels from low through max.
Credit rates remain explicitly unverified where Replit has not published them.
Compile one portable skill
Replit Agent Target
Skill Studio can analyze a workflow, validate it for Replit Agent, and prepare a Replit-compatible configuration while preserving the governed SKILL.md source.
Legacy mode, Turbo, and high-effort fields are migrated with warnings instead of silently rewritten.
Keep the rules with the work
Portable Governance
SKILL.md defines capability, ETHOS.md defines behavior, MEMORY.md defines portable context, and GRAPH.md defines workflow topology. Cromus scores and validates those artifacts before they reach a runtime.
The governance travels; Cromus does not become a runtime lock-in layer.
Review before the merge
CI, Validators & MCP
The Specialist and Cromus Gate can review skills in CI, while free validators and deterministic MCP tools make the same checks available to teams and agent clients.
Model drift checks can flag retired, renamed, or over-provisioned selector choices.
Make trade-offs visible
Cost and Croms Signals
Cromus can surface effort overshoot, inverted model-effort pairings, and model over-provisioning before execution so the team can make a deliberate choice.
Selector credit rates are not invented when the source data is unavailable.
See the selector as a decision surface.
Replit Agent's current selector separates the model from the effort level. Cromus mirrors that shape so a recommendation can explain both axes instead of collapsing everything into a vague "power" label.
- 18 documented selector rows across three modes
- Five effort levels: low, medium, high, extra high, and max
- Unverified credit rates are labeled, never fabricated
Replit Agent selector
Planning view · credit rates unverified
Effort
5 levelsSelected preview: Economy · Medium effort
A practical handoff
From SOP to Replit-ready artifact.
The point is not to automate a setting you should control. The point is to make the decision legible, the artifact portable, and the preflight review repeatable.
Describe the work
Bring an SOP or existing SKILL.md. Start with the real task, not a guess about which model should run it.
Get a recommendation
The advisor and selector signals map complexity, model capability, and effort without pretending to know your account settings.
Validate the artifact
Use Skill Studio, the public validators, CI, or MCP to find governance gaps and migration warnings before the first token is spent.
Choose in Replit
Review the recommendation, then make the final mode, model, and effort choice in Replit Agent. Cromus hands you a portable, auditable artifact.
Clear boundary. Better trust.
Cromus is deliberately upstream of Replit Agent. That boundary is a feature: your team gets a recommendation and an audit trail without handing over control of the runtime.
Cromus can
- Recommend a mode and effort level
- Map model and effort trade-offs
- Validate and score portable specs
- Compile Replit-compatible output
- Flag drift and governance gaps
Replit still controls
- •Your Agent mode and effort setting
- •The models available in your account
- •Platform routing and provider behavior
- •Your workspace permissions and billing
- •What executes after you approve the work
Make the next Replit build easier to defend.
Start with a real workflow. Cromus will show you the assumptions, governance gaps, and model-effort choices before they become runtime surprises.