workflow-runner

Yes, you can run multi-agent YAML workflows inside Claude Code without an API key. The current session LLM acts as the execution engine, reading role definitions and running steps in topological order directly within the tool.

To execute a YAML workflow in Cursor, provide the YAML workflow file and a role directory like agency-agents-zh. The workflow runner parses role definitions, resolves dependencies, renders task templates, and executes steps in topological order.

Topological execution for multi-agent orchestration means resolving task dependencies to run serial and parallel steps in correct order. It ensures each step's output is passed backward to subsequent steps, enabling structured cross-role collaboration within a single conversation session.

Yes, the workflow runner supports parallel task execution across multiple agent roles. It reads YAML workflow definitions, locates role definitions, and executes both serial and parallel tasks while passing each step's output to subsequent steps.

Before running agency-orchestrator workflows in OpenClaw, you need a YAML workflow file, a role directory such as agency-agents-zh for role definitions, and an active LLM session to act as the execution engine for template rendering and step execution.

Use YAML workflow orchestration instead of single-prompt interactions when you require structured multi-role collaboration for product requirement reviews, design reviews, or task decomposition that need dependency resolution and layered execution plans.