I’ve been orchestrating AI agents for a while. However, when the orchestration relies only on prompts or skills, the results can easily drift off track, and it can also consume a lot of tokens. What’s even more troublesome is that these workflows are often hard to edit, maintain, or reuse later.
Before building OpenFlow CLI, I had actually made several versions of similar CLI tools. They were all trying to solve the same problem: making agent workflows more stable, easier to rerun, and easier to maintain.
This time, inspired by Claude’s dynamic workflow idea, I found that writing workflows directly as scripts makes everything much clearer. You can define each step, what each agent should do, and which tasks should run in parallel, all inside a workflow script.
That’s why I built *OpenFlow CLI*.
The benefit of making it a CLI is that it can support different agent platforms, such as OpenClaw and Codex. A completed workflow script can also be executed on its own as an automation workflows.
Personally, I still mainly use Codex. I also have a Gemini annual plan from before. While Gemini may not be everyone’s first choice for every use case, I still find Gemini Flash very useful for simple, well-planned tasks, and its quota is quite generous.
So OpenFlow CLI supports mixing different providers in the same workflow. For example:
* Use Codex for correctness or security reviews
* Use Gemini Flash for testing, organization, and summarization tasks
* Combine the results from multiple agents into a final report
I’ll also look for time to add support for more tools later, such as Antigravity CLI or GitHub Copilot CLI.
The usage is simple. The package includes an `openflow-workflow-writer` skill. Once installed, you can use it to generate workflows or directly run a workflow script:
```bash
npx @prmflow/openflow run workflow.js
```
I personally really like this workflow CLI tool, and I’ve already been using it in some of my daily development work.
If you also find it useful, I’d really appreciate it if you could give my GitHub repo a star. It helps me know that this project is worth spending more time maintaining.
travisliu•1h ago
Before building OpenFlow CLI, I had actually made several versions of similar CLI tools. They were all trying to solve the same problem: making agent workflows more stable, easier to rerun, and easier to maintain.
This time, inspired by Claude’s dynamic workflow idea, I found that writing workflows directly as scripts makes everything much clearer. You can define each step, what each agent should do, and which tasks should run in parallel, all inside a workflow script.
That’s why I built *OpenFlow CLI*.
The benefit of making it a CLI is that it can support different agent platforms, such as OpenClaw and Codex. A completed workflow script can also be executed on its own as an automation workflows.
Personally, I still mainly use Codex. I also have a Gemini annual plan from before. While Gemini may not be everyone’s first choice for every use case, I still find Gemini Flash very useful for simple, well-planned tasks, and its quota is quite generous.
So OpenFlow CLI supports mixing different providers in the same workflow. For example:
* Use Codex for correctness or security reviews * Use Gemini Flash for testing, organization, and summarization tasks * Combine the results from multiple agents into a final report
I’ll also look for time to add support for more tools later, such as Antigravity CLI or GitHub Copilot CLI.
The usage is simple. The package includes an `openflow-workflow-writer` skill. Once installed, you can use it to generate workflows or directly run a workflow script:
```bash npx @prmflow/openflow run workflow.js ```
I personally really like this workflow CLI tool, and I’ve already been using it in some of my daily development work.
If you also find it useful, I’d really appreciate it if you could give my GitHub repo a star. It helps me know that this project is worth spending more time maintaining.