AI Agent Workflow Orchestrator

What problem it solves

Many AI agents can already generate SQL or call individual skills, but turning several steps into a reliable pipeline still requires hand-written DAGs, serial waiting, ad hoc branching, and external command handling. Agent Workflow Orchestrator targets this gap: a single-file, zero-dependency workflow engine for data input, filtering, Join, aggregation, Python, CLI, and Skill nodes, with sequential, parallel, and auto execution modes.

How it works

The workflow is defined by JSON-style node declarations. Data nodes cover DataIn, Filter, Join, Aggregate, and DataOut; control nodes include Branch for conditional routing and Python / CLI for extension. auto mode infers the execution shape from the DAG topology: serial chains stay sequential, independent branches can run in parallel, and conditional branches route by upstream output. Skill calls can bring in OpenClaw skills such as smart-data-analyzer, duckdb-data-analysis, and mysql-analytics, so analysis, visualization, and reporting steps can live in the same workflow. The runtime also enforces safety guardrails, including SQL allowlists, path validation, and limits on execution time, node count, file size, and memory.

Boundaries

This is best viewed as a lightweight data-processing pipeline and agent task orchestrator, not a replacement for enterprise schedulers like Airflow or DataWorks. It is not the right choice when workflows require distributed scheduling, complex retries, or large-scale cluster resource management. When using Python, CLI, or external skills, review input paths, command arguments, and SQL permissions before exposing them to untrusted data.