@zeqqat/workflow-engine

The executor for workflow definitions: templating, retries, polling, branching, parallelism, saga compensation, approval gates and redaction of sensitive outputs.

It performs no I/O of its own. It does not know what a database is, or an HTTP client. Step types are supplied by you through a HandlerRegistry, and everything the run emits travels through an injected ExecutorEmitter. That seam is why the same executor writes Postgres rows in a server and SQLite rows in a CLI without being modified.

Part of Zeqqat, which exposes declared workflows as MCP tools. Published separately so the @zeqqat/mcp CLI can depend on it directly.

Install

npm install @zeqqat/workflow-engine

Use

import { WorkflowExecutor, HandlerRegistry } from "@zeqqat/workflow-engine";

const registry = new HandlerRegistry();
registry.register("http", myHttpHandler);

const executor = new WorkflowExecutor(registry, emitter);
const result = await executor.execute(workflow, inputs, callerContext);

A handler receives a step's resolved inputs and returns its declared outputs. The executor owns everything around that: what runs next, what retries, what gets rolled back.

Compensation

When a step fails, completed steps are compensated in reverse order, and the result says whether the world was left clean. That distinction — "this failed and nothing happened" versus "this failed and something is half done" — is the point of the saga support, because it is what tells a caller whether retrying is safe.

License

Apache-2.0