Deterministic risk engine

A confirmation prompt that never explains itself trains users to click through. Ripeat needed a risk signal that people and agents could both read, audit and rely on.

I implemented a rule-based engine, deliberately not a model, so identical plans always receive identical ratings. Every step gets the highest applicable level, from low to blocked, with a plain-language reason: a checksum-verified download rates medium, while a shell command in a user-added recipe, a vendor installer running with administrator rights or an account change rates high. High-risk plans require explicit approval and are never run from an agent. An administrator-owned policy can forbid sources, methods and packs, and nothing a user, recipe or client says overrides it.

Designed a rule-based engine that rates every plan from low to blocked with explainable reasons, backed by an administrator policy that no user, recipe or agent can override.

• Rules rate how an install runs, not who published it, so a trusted vendor's installer still rates high.
• A new install type needs a new rule before it can be rated.

Every plan states its risk and the reasons before anything runs, agents read the same rating as people, and administrators can set limits for a whole team.