Hi. I’m the author of backlog.md . Your team did a great job on Vibe Kanban. It looks great and has a cool name. I like to see more tools trying to solve the human-ai agent interaction for building projects.
Backlog.md is still a bit rough on the edges but is definitely a proof that markdown files and AI Agents work really well together. I started working on it exactly a month ago
It is so confusing that I had to add custom instructions for Codex telling him that Backlog.md is the project folder and not a file. He was wasting few mins trying to CAT Backlog.md instead of CD
Correct. One of the instructions that ships with backlog.md is to make the tasks “as big as they would fit in a pr”. I know this is very subjective but Claude really gets much better because of this.
1) How will it work with multiple branches?
Simple: using git :)
Git allows to fetch certain files from other branches including remote ones without checking out those branches.
The state is always up to date no matter if you are running backlog.md from main branch or a feature branch.
It works well when there are not many branches but I need to check if I can improve the performance when there are lots of branches.
I sent a message to someone telling that I was working on backlog.md and it turned the name into a link automatically.
I wanted to remove the link and I clicked on it accidentally and discovered that not only there was nothing on that domain but was not registered yet. I got the domain few mins later :)
When you initialize backlog in a folder it asks you if you want to set up agent’s instructions like CLAUDE.md. It is important to say yes here so that Claude knows how to use Backlog.md.
Afterwards you can just write something like: Claude please have a look at the @prd.md file and use ultrathink to create relevant tasks to implement it. Make sure you correctly identify dependencies between tasks and use sub tasks when necessary.
Or you can just paste your feature request directly without using extra files.
Will definitely do.
I am also planning to run a benchmark with various models to see which one is more effective at building a full product starting from a PRD and using backlog for managing tasks
I threw Claude Code at an existing codebase a few months back and quickly quit—
untangling its output was slower than writing from scratch. The fix turned out
to be process, not model horsepower.
Iteration timeline
==================
• 50 % task success - added README.md + CLAUDE.md so the model knew the project.
• 75 % - wrote one markdown file per task; Codex plans, Claude codes.
• 95 %+ - built Backlog.md, a CLI that turns a high-level spec into those task files automatically (yes, using Claude/Codex to build the tool).
Three step loop that works for me
1. Generate tasks - Codex / Claude Opus → self-review.
2. Generate plan - same agent, “plan” mode → tweak if needed.
Should you use one, the other or both?