I recently added a simple task-management layer to my AI CLI setup, and it has been a big quality-of-life upgrade.
The core idea is straightforward: keep work trackable without adding process overhead.
- Tasks use stable IDs (
0001,0002, ...) - Bugs use their own stable IDs (
BUG-0001,BUG-0002, ...) - Active files stay small (
TODO.md,BUGS.md) - History is preserved (
DONE.md,BUGS_DONE.md,REMOVED.md) - Progress is append-only (
progress_log.md)
What I like most is that it works for both solo and agent-driven workflows. I can run tasks in sequence or parallel, and still keep clear visibility on what’s active, what’s done, and what was dropped.
I also added optional Discord notifications for agent progress. Secrets are handled safely by default .
The goal wasn’t to build another project-management system. It was to keep execution resumable, context small, and momentum high.
If you’re using coding agents, I highly recommend having a minimal, file-based task protocol like this. It scales surprisingly well.
Feel free to check it out in my repo dedicated to agent skills, policies and configurations: https://github.com/aleksandarristic/ai-cli-config
Comments