---
name: xentropy-workflows
description: Create service simulation agents, compose workflow DAGs, and run end-to-end tests locally with xEntropy.
---

# xEntropy workflows

Use xEntropy to replace unavailable service calls with agent simulations and test connected workflows. Read [the API reference](references/api.md). The platform's authoritative design is `/docs/service-simulation-platform`.

Create service definitions with behavior instructions, selected context files, operations and initial state. Add graph nodes that explicitly choose real or simulated execution. Connect dependencies and nested output bindings. Add separate test cases; keep expected outputs in assertions, never in agent context.

Run the graph in the user's browser or local Node.js runner. Never upload the complete workflow for sandbox execution. One run is one session; each test case gets fresh simulated state. Real services are not reset automatically. Use a scoped runner token supplied by the user; keep it outside graph files. Resolve real-service secret headers from environment references locally.

Download `/resources/xentropy-runner.mjs` or use `npm run graph:run -- --help` in the repository. Inspect its help before use. Long-running simulated requests are polled by ID; matching retries must retain their idempotency key. Do not retry uncertain real side effects. Report actual run results and distinguish agent failures from service outputs.

Never claim recorded examples or injected responses are live simulation evidence. The old Prepare/Run command-workspace interface and Python helper are retired; there is no fallback to them.