af-orchestrate

Workflow orchestration evaluates the current state, progress, and last step outputs to output a deterministic JSON decision guiding the next action. This enables autonomous progression, conditional evaluation, and robust error recovery in multi-step tasks.

AI agents handle retry logic by applying conditional evaluation to the workflow state and contextual data. They return a deterministic JSON decision that instructs the agent whether to retry, pause, or advance the workflow sequence.

Yes, you can define workflow structures in YAML for conditional evaluation. The orchestrator evaluates the YAML alongside progress JSON to output deterministic decisions for blocking conditions and human-in-the-loop scenarios.

The best way to keep workflows moving autonomously is to use a decision engine that predicts next actions by evaluating workflow state. It outputs a JSON decision format that agents execute to advance, pause, or complete tasks without manual intervention.

You do not strictly need Nunjucks templates for decision reasoning, but the orchestrator supports Nunjucks for contextual data formatting. It evaluates the workflow state and outputs deterministic JSON decisions regardless of the templating approach.

You should avoid using a JSON decision engine for workflow orchestration when tasks require non-deterministic outputs or lack strict conditional evaluation requirements. It is designed for robust decision reasoning in multi-step automation, not simple linear scripts.