azoth-eval-swarm

Multi-wave swarm evaluations are coordinated by routing prompts through an eval-swarm workflow that preserves stage structures, gate rules, and agent binding to an orchestrator across iterations. The workflow reads command.yaml as truth and loads the body from the referenced contract to execute strict evaluation scenarios.

A strict swarm evaluation scenario is a process defined by a command contract that enforces evaluation rules, stage structures, gate rules, and agent binding. It uses isolated evaluators and multi-wave iterations to assess prompts while honoring the azoth_effect contract for consistent orchestration.

Yes, custom arguments are supported by treating trailing prompt content after the invocation command as $ARGUMENTS for the workflow. This flexible argument handling allows you to inject specific parameters into the eval-swarm workflow during prompt routing.

The eval-swarm command contract structures evaluation stages by reading command.yaml as the source of truth and loading the body from a referenced markdown file. It preserves stage structures, gate rules, evaluation rules, and agent binding across all interactions to maintain strict evaluation integrity.

Yes, an orchestrator agent binding is required to coordinate isolated evaluators in a swarm. The workflow enforces this agent binding alongside gate rules and evaluation rules to ensure strict multi-wave iterations execute within the defined command contract.

Swarm evaluation fails if the command contract is not honored because the workflow depends on reading command.yaml as truth and preserving stage structures, gate rules, and the azoth_effect contract. Ignoring these constraints breaks agent binding and disrupts isolated evaluator coordination across multi-wave iterations.