council

You can orchestrate multiple AI models for plan review by running parallel review sessions via pi, using centralized session state to track iterative critique until consensus is reached. This coordinates architecture decisions and risk assessments through per-agent prompts.

The best way to conduct an iterative risk assessment with AI agents is using a session-based workflow that coordinates parallel reviewers. This approach enforces deterministic iterations via a structured decision log, allowing continuous refinement until consensus is achieved.

Yes, you can inspect specific files during a multi-model review by passing per-review --ref paths and providing guidance on what to inspect. This selective file inspection allows targeted architecture decisions and quality audits.

To start a session-based plan review, run council setup to verify dependencies, then execute council review plan.md --prompt "Review this plan for gaps and risks". Continue iterations using council review --session <session-id>.

Yes, multi-model plan review requires a structured decision log to enforce deterministic iterations during the session-based workflow. This log preserves session state and tracks iterative critique across parallel AI reviewers.

You should avoid using parallel AI reviewers when manual, fragmented plan reviews are sufficient or when session-based iterative refinement is unnecessary. The approach is designed for complex architecture decisions, risk assessments, and quality audits requiring multi-model consensus.