workflow-runner

Multi-step agent pipelines are automated by parsing TOML workflow definitions that specify explicit steps, dependencies, and inputs. The workflow runner validates skill availability, enforces acyclic dependency order, and executes named skills sequentially to coordinate complex automation tasks reliably.

Independent workflow steps run concurrently when the TOML workflow definition specifies them without mutual dependencies. The execution engine resolves the dependency graph and parallelizes independent steps, returning a structured aggregated result with per-step statuses.

TOML workflow files define dependencies and inputs by explicitly listing named skills as steps and mapping outputs from prior steps to dependent steps. The runner validates skill availability, detects circular dependencies, and propagates outputs between steps automatically.

Workflow execution halts immediately on step failure rather than skipping to the next task. The runner stops the pipeline and returns clear failure reasons alongside the aggregated result, ensuring strict safety for CI-like automation and headless pipelines.

TOML workflows are designed for CI/CD style pipelines, multi-agent code generation flows, and headless automation tasks. They enforce deterministic execution, validate skill availability, and aggregate per-step outputs into a single structured result.

A circular dependency error occurs when the TOML workflow definition contains steps that recursively depend on each other. The runner enforces acyclic dependencies to ensure safe execution, halting the pipeline and returning a clear failure reason when cycles are detected.