Will llxprt-code be ZCF-compatible?

[BradKML]

Take a look at this toolset (ain't no Claude Flow but close enough as a one-stop shop) https://github.com/UfoMiao/zcf

[acoliver]

If they support gemini-cli they could probably support llxprt-code. There are several of these like: https://github.com/Theopsguide/gemini-code-flow

Right now ZCF seems to be hard coded to claude code unless I misunderstood something. It probably would be easier to update it to call different CLIs such as llpxrt-code. I'm probably not going to change the CLI to be executable with "claude" and change all the command line options to match claude (as the features of this tool support more things and require more info).

With that said, it probably wouldn't be hard to write a shell script and alias it to "claude" and have it call it. Not exectly zero config but close. If its calling non-interactive mode for claude anyhow or integrating with MCPs -- all of that is doable. So I'm guessing you could use it!

In the long term, for LLxprt-Code, we'll support natively SOME of the features of Claude Flow. I'm not convinced the FANN network, which is only partially implemented, actually is providing substantial value. I also do not think it is useful to launch a swarm of 30 agents, especially since the Claude Flow people destroyed Anthropic's largess and put us back on limits. However, I do think the SPARC idea has a lot of merit, but needs more structure ( tools that create it in a more deterministic way with goals and objectives that are easily measured) and more hardboiled execution (on RooCode SPARC agents blow through the test gates, and you still get stub code). I also think having multiple models is more sustainable as Anthropic continues to close the doors and dumb down their models, as experienced in Claude code.

So short term: shell script and an alias and I think you can make it work.
Long term: some of the ideas from Claude flow without the extravagance (i.e., more like 2-3 agents and only parallel if it's truly a parallel task) and SPARC with more control (more structure and goals in the plans and hard-boiled test and quality gates) and more models, including local. And also some different ideas like https://www.reddit.com/r/godot/comments/xgrk0g/goap_goaloriented_action_planning_is_absolutely/ and other forms of adversarial review and static analysis.

Working on multi-model Subagents right now.

[BradKML]

How many other tools have you seen with subagents? Cursor Agents seems to be lacking, so I am kinda trying to think laterally about how it is possible to run Gemini-CLI or Claude Code in Cursor/Windsurf and Roo/Kilo https://github.com/eastlondoner/vibe-tools

[acoliver]

Think linearly and temporaly into next week. We'll have subagents by then. Sooner if you like testing my worse bugs. Im literally working on this atm.