workflow-runner

You can execute multi-role agent workflows by loading a YAML file into your current session. The session's LLM acts as the execution engine directly, parsing agents, inputs, and steps without requiring any external API key configuration.

YAML workflow orchestration for AI agents defines names, agent directories, inputs, and execution steps in a YAML file. The orchestrator reads role prompts from predefined directories and executes tasks in topological order using the current LLM session.

Yes, Cursor can execute agency-orchestrator YAML files directly. The workflow runner supports AI tools including Cursor, Claude Code, and OpenClaw, using the active session to render step tasks with prior outputs and enforce a fail-safe execution plan.

Previous step outputs are handled by rendering them into the task text of subsequent steps. The orchestrator propagates context by combining input variables with prior results, storing per-step results in context for layered parallel or serial execution.

No, you do not need to install external dependencies to orchestrate agents. The workflow runner operates entirely within the current AI tool session, reading role prompts and executing YAML workflows using the built-in LLM without extra packages.

Limitations of running YAML workflows inside AI tools include reliance on the current session's LLM for execution capacity and context window size. Complex multi-role collaboration workflows are constrained by the session's token limits and the need for predefined agent directories.