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.
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've had same experience. Taskmaster-ai was pretty good, but sometimes the agent ignored it as the project grew larger (can probably prevent that now using claude code hooks).
Trying this project today looks nice. I see you have sub-tasks. Any thoughts on a 'dependency' relation? I.e., don't do X if it is dependent on task A which is not complete.
FYI, there is a 404 in the AGENTS.md GEMINI.md etc pointing to a non existing README.md.
Would love more detail on your integration with claude. Are you telling claude to use backlog to plan X task? Feels like some MCP integration or something might make it feel more native?
Though i've not had much luck in getting Claude to natively use MCPs, so maybe that's off base heh.
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.
With aider you can run a second instance along with --watch-files and if you in your tasks do // #AI it will be added to the chat and with // !AI AI will then respond
so you can do
`backlog task create "Feature" --plan "1. Research\n2. Suggest Implementation// #AI AI!"` (yes weird order with the !)
and in the background aider will propose solutions.
I’m not sure how this compares to Claude Code or Codex, but its LLM-flexible. Downside is it doesn’t create a pull request. So it’s more helpful for local code.
I would probably add some Readme.md files to the --watch-files session and I think you need to click once [D]ont ask again so it wont keep asking you to add files
You can use this perfectly fine without AI agents, it just so happens to produce output which is easily ingestable by LLMs. It also has drag and drop visualisation and simple syntax for creating and tracking tasks in your codebase.
I built myself a tool that does something quite similar. It's a single no-dependency Python script that parses "tasks.md" in the root of the repo which contains a Markdown table of tasks, then has basic support for BLOCKED/READY/DONE/CANCELLED, dependencies, titles, tags, etc.
For a project that is just for me, it's exactly what I need – dependency tracking and not much more, stored offline with the code. Almost all of the code for it was written by Gemini.
This is a good idea. But the screenshots you have show lots of tasks in a project; how are you dispatching tasks (once planned) to an agent, and how are agents navigating the large number of markdown task content you're producing without blowing out their context budget?
Thanks for backlog.md. We love it for our project. The problem we face now is that we have two separate repo for BE (Golang) and FE (next.js). And we wanted to have one backlog for both projects. But backlog resides inside git repo.
Seems like a great idea. How would that work with multiple branches ? One task might be implemented in a different branch, we might want to have a global overview of all the tasks being coded in the main branch
All data is saved under backlog folder as human‑readable Markdown with the following format task-<task-id> - <task-title>.md (e.g. task-12 - Fix typo.md).
If every "task" is one .md file, I believe AI have issues editing big files, it can't easily append text to a big file due to context window, we need to force a workaround launching a command line to append text instead of editing a file. So this means the tasks have to remain small, or we have to avoid putting too much information in each task.
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.
Nice, so there could be some kind of git kung fu command line to help with that. Maybe we could also have a separate folder using git worktree to post all the information in one branch. That'd duplicates all files though.
Readit News
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.
3. Implement - Claude Sonnet / Codex → review & merge.
For simple features I can even run this from my phone: ChatGPT app (Codex) → GitHub app → ChatGPT app → GitHub merge.
Repo: https://github.com/MrLesk/Backlog.md
Would love feedback and happy to answer questions!
Would love to see an actual end to end example video of you creating, planning, and implementing a task using your preferred models and apps.
I look forward to a future where we are reduced to rubberstamping fully-agentic-generated code on our glass slates for $0.01 eurodollars a PR.
Trying this project today looks nice. I see you have sub-tasks. Any thoughts on a 'dependency' relation? I.e., don't do X if it is dependent on task A which is not complete.
FYI, there is a 404 in the AGENTS.md GEMINI.md etc pointing to a non existing README.md.
Will check the 404 issues. Thanks for reporting it
Though i've not had much luck in getting Claude to natively use MCPs, so maybe that's off base heh.
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.
Feels a bit like magic
so you can do
`backlog task create "Feature" --plan "1. Research\n2. Suggest Implementation// #AI AI!"` (yes weird order with the !)
and in the background aider will propose solutions.
I’m not sure how this compares to Claude Code or Codex, but its LLM-flexible. Downside is it doesn’t create a pull request. So it’s more helpful for local code.
I would probably add some Readme.md files to the --watch-files session and I think you need to click once [D]ont ask again so it wont keep asking you to add files
The tagline from the repo seems fine: "A tool for managing project collaboration between humans and AI Agents in a git ecosystem"
> Rich query commands -- view, list, filter, or archive tasks with ease
If these things appeal to you and you haven’t already looked at it, the GitHub CLI tool gh is very useful. For instance:
— https://cli.github.comYou can do things like fork repos, open pull requests from your current branch, etc.
For a project that is just for me, it's exactly what I need – dependency tracking and not much more, stored offline with the code. Almost all of the code for it was written by Gemini.
Because of the embedded custom instructions Claude knows exactly how to proceed.
Since I never create too big tasks, what blows most context are actually the docs and the decisions markdown files.
Can you propose a neat solution of this problem?
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.
https://github.com/MrLesk/Backlog.md/blob/main/src/guideline...
You will notice yourself that smaller atomic tasks are the only way for the moment to achieve a high success rate.
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.
Another idea is to use git notes