Run two AI agents in parallel to review a pull request from different angles, then pipe both outputs into a third agent that writes a combined summary.
Break a large refactor task into sequential AI coding steps, where each step receives the previous step's output as context.
Test a multi-agent workflow locally using the built-in mock provider before running it against real AI API calls.
Requires the AI CLI tools you want to run (Codex CLI, Gemini CLI) to be installed and authenticated separately before workflows can execute.
OpenFlow is a command-line tool for running multi-step workflows that involve AI coding agents. There are several AI coding assistants available as terminal commands, such as Codex and Gemini, and OpenFlow sits above them as a coordinator. Instead of running each agent manually one at a time, you write a workflow file that describes what should happen and in what order, then OpenFlow handles the execution. The main use case is breaking a large coding task into stages. For example, you might want two different AI agents to review a pull request in parallel, each looking at different concerns, and then have a third agent summarize both reviews. OpenFlow lets you write that sequence as a TypeScript file and run it with a single command. It tracks which steps finished, captures the output from each agent, and saves a record of the entire run to a local folder so you can inspect what happened. Workflows are written in a constrained subset of TypeScript using a small set of built-in functions: one for running a single agent task, one for running multiple tasks at the same time, one for running tasks in a pipeline, and one for marking named phases. The tool validates workflows before running them, checking that the structure follows its rules and that required metadata like a name and description are present. OpenFlow does not include its own AI model. It calls out to whichever provider CLI tools you have installed locally, such as the Codex CLI or the Gemini CLI. A mock provider is included for testing workflows without making real API calls. The tool is available as an npm package and can be run without a permanent installation using the npx command. It supports options for controlling how many tasks run at the same time, setting timeouts, choosing an output format for use in automated pipelines, and stopping the whole workflow if any single step fails.
← travisliu on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.