I forked it and added a bunch of stuff on mine...

[jsh562]

+extended and plugged with Deepseek V4 full support (except images, and web search is still not done), Minimax M3 half done, Claude updated
+Full strict process mode toggle; and Planner, Worker, Reviewer, QC, Integrator roles per project (with full integrated pipeline support with permissions, etc.) that can be added or removed from any agent (including god). so if toggled, it uses all those roles, if not toggled it uses the original set.
+Task board full strict integration with process and agents workflow, with a separate view on the left window pane
+triage tasks function (fix/reassign assignees, blocked tasks, etc)
+Project Repos+Workspaces+Worktrees+Home Directory working together without overlap
+Project Repos/Phase agent status view

on top of your v0.1 i think.
but I made a mess of the commits... do u want any of these things? lol

[chaitanyagiri]

Hey that'd be wonderful if you can raise a PR here I'll review

[chaitanyagiri]

I have also added a bunch of features recently, we now support most cli agents even opensource ones

[jsh562]

Sorry I just got back to this.

I did a high level diff analysis between what you have currently from Today vs what I have

Your implementation of Deepseek is very different from my implementation.
Your implementation runs the agent as a subprocess as a small addition to the existing structure where the CLI does everything and the app watches the terminal outputs and hooks.
My implementation runs the deepseek agent differently where it is more integrated into the app. The agent loop runs in-process and calls the LLM api directly and the agent executes tool calls in-process, parses the stream, etc.
They also have read_file/edit_file/grep/bash/web_search + spawn_subagent + hive_* commands + git worktrees directly thru the app in-process, and each agent makes changes in their own worktrees and coordinates with each other thru the app instead of like thru the CLI.
I also separated out the deepseek agent-core into its own package.

also, I tried to separate out the features so it might be easier to submit each individually... like the Kanban features vs the other features but all of the features are so intertwined with each other I don't think I can really separate it easily. To do it like that, I would have to make each feature be able to handle missing sister features well before I can ship them individually.

I think there's a way to layer the features depending on their dependencies.... I will do some thinking lol