ralplan

Multi-agent consensus planning works by orchestrating a Planner, Architect, and Critic in iterative loops. The Planner generates a plan, the Architect evaluates soundness, and the Critic validates testability, looping up to five iterations until agreement is reached.

To create an ADR-style plan for high-risk architecture reviews, invoke a deliberate planning workflow. This adds pre-mortem analysis and expanded test planning to the standard consensus loop, producing a bounded plan with formal decision records for approvals.

Yes, consensus planning supports both interactive and non-interactive workflows. You can use prompts that require user input for manual guidance or run fully automated planning, with both modes producing final ADR-style artifacts for traceability.

The best way to reduce ambiguity and risk in feasibility assessments is using a 3-actor planning loop. A Critic validates quality while an Architect provides steelman arguments, iteratively refining the plan until consensus is achieved and a bounded plan is produced.

A key limitation of automated consensus planning is the iteration cap; the re-review loop runs a maximum of five times. If consensus is not reached within these bounded iterations, the workflow may not produce a finalized, accepted plan.