HN Explorer

Show HN: Voro – An attention manager for agentic coding (github.com)

1 points by ClachDev ↗ HN
A few months ago I was getting frustrated trying to use Claude Code effectively. I was finding that for small tasks I would spend a lot of time waiting for it to finish "thinking" and I would end up browsing here...

To make a better use of my time, and ideally make better use of the limits, I developed Voro to help me know what I should be focusing on.

Fundamentally it's a task tracker tailored for local agentic coding. Each task is a plan that gets a priority and works its way through a set of states that are either for "humans" or "agents". The main "cockpit" view lists a few different options of what I should be looking at. The bottom has a list of what your agents are working on.

Agents are given a small description and generate a task. That can then be reviewed and accepted if it's good, or iterated on by refining. Once accepted it can be dispatched to an agent for implementation. Once finished it comes back for review which can again then be iterated on. The crucial thing is that many tasks can be in flight at once. While I'm waiting on a plan being refined I can be reviewing the output of another one.

This is still an early project but I've been using it for a few projects I'm working on, both new codebases and large existing ones. It's probably overly tailored for my workflow right now and I haven't tested it with anything but Claude Code. I'm curious if anyone else has built something similar? Please let me know what you think!

2 comments

[ 5.3 ms ] story [ 12.0 ms ] thread
fottly
Nice, this scratches an itch I've had for a while. Curious how you're handling attention/context persistence across sessions, local only, or synced somewhere?
ClachDev
It makes sense that I build Voro using itself which highlighted some issues with it.

Firstly, some "human" tasks get stuck at the top of the priority list because the effort involved seems large relative to agent dispatched tasks. I feel like this is a common scheduling problem?

Secondly, its sometimes hard to judge progress. When the code moves fast, it's impossible to keep it all in your mind at once. Instead I found myself focusing on the plan and making sure it was what I actually wanted and then going lighter on the review of the code, testing the desired functionality.

I feel like agentic coding is like learning software engineering all over again. How do we take advantage of AI to write the wrote bits quickly, but keep the quality?