I've been abusing Claude Code for the last few months; however, something that always slowed me down is task management, a real one.
I built Swarmit for a simple objective: ask Claude to plan and organize itself using a proper task management system. Since building it, it creates tens of epics and hundreds of tasks, and has been able to reliably use them to manage parallel agents, manage dependencies among tasks, and, most importantly, not forget about what it was supposed to do.
Why not GitHub/Jira/Asana? This lets you do everything locally. It’s meant to allow Claude to quickly organize itself without reaching external services or polluting your official issue tracking with AI-generated micro-tasks.
I added a simple TUI that allows you to quickly see what the agents are doing, to comment and interact with the tasks. Started working on Insights so that agents can add some gotchas.
I hope you'll enjoy using it as much as I do :)
PS: 100% dogfooded, and my first fully claude code built tool.
The ones I use:
- python_parser, allows me to load the python source code and detect any parsing errors (returned to the frontend)
- python_ast, allows me to extract the identifiers (variable names, function names, etc)
- python_semantic, allows to do scope analysis and understand relationships between identifiers
In the first prototype I used Python's ast and symtable libs. However, I really wanted to rely on the speed of Ruff's implementation and Rust's speed.
Thanks for the feedback. You're right, it's part of the long term goals.
I'd like to keep the center of the screen clutter free and add side-bars later on with tooling. Among them, an inspector just like what PyCharm has when opening a notebook. For globals it's quite simple to add, however to inspect the content of a single compound statement it might be a bit tricky (if you've got ideas I'd love to learn more).
Some of the next "urgent" features are improving the code state management (history), having multiple sessions, being able to save them, etc. Maybe adding a way to bring the data online somehow.
I noticed that redshift stopped working lately due to geoclue unable to get any location, it by default uses a mozilla location service which is failing lately with a 403.
My solution is to set manually the lat/lon in redshift configuration.
No typo there. When we talk about vectors we mean "column vector". As it's easier to read horizontally (and takes less place in a paper), most of the time we write x^T = {a, b, c} rather than writing them in a column shape.
I've been abusing Claude Code for the last few months; however, something that always slowed me down is task management, a real one.
I built Swarmit for a simple objective: ask Claude to plan and organize itself using a proper task management system. Since building it, it creates tens of epics and hundreds of tasks, and has been able to reliably use them to manage parallel agents, manage dependencies among tasks, and, most importantly, not forget about what it was supposed to do.
Why not GitHub/Jira/Asana? This lets you do everything locally. It’s meant to allow Claude to quickly organize itself without reaching external services or polluting your official issue tracking with AI-generated micro-tasks.
I added a simple TUI that allows you to quickly see what the agents are doing, to comment and interact with the tasks. Started working on Insights so that agents can add some gotchas.
I hope you'll enjoy using it as much as I do :)
PS: 100% dogfooded, and my first fully claude code built tool.