149Build a graph-based workflow editor with cycle detection and execution ordering

Tools like Zapier, n8n and GitHub Actions let users connect steps into a directed acyclic graph (DAG): "when a form is submitted → enrich the lead → (send email, add to CRM) → notify Slack". Implement the model behind such an editor:

createWorkflow() returns:

• addNode(id, run): run(inputs) is async; inputs maps each predecessor id to its output. Duplicate ids throw.
• addEdge(from, to): throws for unknown nodes or self-loops, ignores duplicates, and throws Error("Edge a -> b would create a cycle: b -> c -> a") (the path from to back to from) instead of adding an edge that creates a cycle.
• wouldCreateCycle(from, to): the same check without changing anything (the UI uses it to show a red line while dragging).
• removeEdge(from, to), removeNode(id) (also removes its edges).
• order(): a topological order; among nodes that are ready at the same time, keep the order in which they were added.
• levels(): groups of nodes that can run in parallel (level 0 has no predecessors, level n depends only on earlier levels).
• run(): execute the workflow. Each node starts as soon as all its predecessors succeeded; independent nodes run concurrently. If a node fails, record its error and skip everything downstream of it, but keep running independent branches. Resolve { results, errors, skipped } (results and errors keyed by id, skipped a sorted list).