General Direction for the Project/Task Management side of things

[PeteHalsted]

@coleam00 and gang,

I know you guys have a lot on your plate and some big plans for Archon, but I was hoping you could share some kind of road map or at least big picture plan. There may be more information in the "community" but that is pay walled off and keeps the general public out of the loop which I fear might prevent Archon from getting the traction it deserves.

We are evaluating it now to see if we want to add it to our workflow and I of course understand it is early on, and I am not expecting everything to be there but it would be nice to have an idea of what "might" be there.

I know you have put a lot of effort, and continue to, the knowledge base/RAG system, and it is impressive as I already shared in other post. And after playing with Archon for a few days I am really getting a vision for what it could be, and how it would be exactly what we want/need if it were. I could fork the project and create some of this but lets be honest I have enough rabbit holes already, and I would much rather know it matches the direction of the core project than go rowing down a different stream.

So I am going to describe my thoughts on what I think it could be and I would love some feedback on if any of this sounds like a direction you will be going.

Note: I talk about Claude below but most of this could work with other coding agents as well.

1. Right now Claude "drives" Archon. That is backwards, to achieve what I want, Archon needs to drive Claude. Which is pretty easy to do with the SDK, headless mode, etc. Lots of examples of other projects out there that do this.
2. Now that I have Archon driving Claude (see what I did there always assume a yes!) , we have the ability for some powerful features. That will make Archon our development command center.
3. BTW, we still need 2 way communication, claude/gemini etc. planning and then pushing the task to Archon is great. Not just TodoWrite but actual spec task.
4. So with all that said, I now have great task in Archon.

• I can assign them to an Agent. Not just Claude but a specific agent. Be that an actual Claude Subagent or via prompt passing "You are a ...."
• I can configure that Agent with specific tools/MCP. This might be a bit of a stretch I am not sure what can be done as far as custom MCP access per Claude session
• I can link to specific project docs, from the project docs section of Archon. Give the agent just the context it needs to complete this task
• I can link to specific knowledge base entries in Archon for any documentation the agent might need. Again give it the context it needs, so it doesn't need to guess.
• know that I have the task and agent all configure I can launch that task. Which will fire up a claude session in a git worktree, do the work, when done write its summary to a note on the task (not just mark the task done), so I can now review what it said it did as well.
• after review and approval it can then merge the git worktree into the origin.

This would make Archon a one stop Development Command center. Claudia (now called Opcode), Conductor, Flow, etc. all do some parts of this but they don't have the cohesive backbone foundation that Archon has with the knowledge base and project documentation to make it truly efficient.

While you are at it (I always want more!), the project documentation section can also act as a context/memory manager, to replace things like Mem0, Pieces, Byterover, etc. Which would then make it even more powerful when you are able to link those memories to task that you are going to hand off to Claude.

I would be happy to get involved and contribute, joint the community, etc. if this is in anyway the direction/vision you have for Archon to eventually go. I am just looking for some general feedback if my vision is way off from yours and if I need to find a different boat to climb aboard.

Thanks!