advisor-critique-loop

You can automate multi-model critique loops by pairing a Driver model with a Critic model across four project phases. The pipeline iterates through requirements, design, and implementation plans, generating structured artifacts until all critiques are resolved.

A driver/critic review pipeline is a structured process where one model generates outputs and another critiques them. You need it for complex projects requiring iterative refinement to catch edge cases and enforce quality across requirements, design, and implementation phases.

Yes, the advisor critique loop requires claude, codex, and jq as dependencies. These tools enable the multi-model diversity needed for the Driver and Critic roles throughout the gated review phases.

Place a request.md file at your project root and execute run.sh to start the review pipeline. The script orchestrates the driver/critic critiques across the four phases with automatic artifact generation and per-phase validation.

Yes, the multi-model review pipeline includes escalation and iteration controls to prevent infinite loops and safeguard timelines. It enforces per-phase LGTM termination, stopping the critique loop once all issues are resolved.

The best way to validate implementation plans is using a gated review pipeline that enforces model diversity between Driver and Critic roles. This approach applies per-phase validation loops to ensure consistency and catch edge cases across all project phases.