workflow-runner

You can run YAML workflows in Claude Code without an API key by using an in-session runner that parses the YAML, resolves agent directories, and uses the LLM directly as the execution engine for multi-role collaboration.

Multi-role orchestration in YAML workflows involves parsing defined agent roles, building a topological execution plan by DAG layers, rendering role prompts, and executing steps sequentially while preserving outputs for downstream layers.

To trigger multi-role collaboration, simply provide a YAML workflow file in your current session. The runner automatically parses the workflow, locates role directories, gathers required inputs, and assembles the execution plan.

Yes, you can execute agency-orchestrator workflows inside Cursor or OpenClaw. The runner operates entirely in-session without external tooling, enabling seamless multi-role collaboration within a single conversation.

No, you do not need external tools or API keys to parse and execute YAML agent workflows. The runner uses the current session's LLM to parse YAML, render role prompts, and execute DAG layers without any external dependencies.

During YAML workflow execution, the runner collects required inputs before starting and automatically preserves outputs from each step, passing them downstream across DAG layers to ensure sequential multi-role tasks receive the correct data.