critic-gate

To enforce a pre-execution check, route completed AI-generated project plans to a critic agent for mandatory review to block flawed code and wasted development effort. This validates requirement coverage before any implementation work starts.

When a critic review rejects an AI coding plan, a hard stop blocks all implementation. The system requires user input to resolve the fundamental flaws before proceeding to ensure only approved work executes.

You can revise an AI project plan through a structured revision loop supporting up to two rounds. This allows you to refine flagged issues like missing security requirements or logical gaps before implementation begins.

To validate requirement coverage for OpenCode Swarm plan files, trigger a critic review immediately after the architect agent completes the .swarm/plan.md file. This ensures the plan is checked for missing requirements before code is written.

You need an OpenCode Swarm AI coding workflow environment where an architect agent generates .swarm/plan.md files. Gated execution applies to these workflows to intercept new features, bug fixes, or system changes before implementation.