What is a task board for AI coding agents?
A task board for coding agents is a shared list each agent reads before it starts and claims before it works. When a team needs one, and what to look for.
A task board for AI coding agents is a shared list of work that agents read before they start and claim before they write code, so two agents never build the same thing. You need one as soon as more than one agent works on the same repository at the same time, whether they are your own parallel sessions or your teammates' agents. The board only works if claiming is enforced: an agent that tries to take a task someone else holds must be refused and told who has it.
A board for people is a planning tool. A board for agents is a safety mechanism. Agents start fast, never ask "is anyone already on this?" unless told to, and will happily implement the same feature twice in two branches.
What does an agent task board need to do?#
| Need | Why it matters for agents |
|---|---|
| One holder per task | Two agents reading "top item: add login" at the same minute both start it |
| A refused second claim that names the holder | The second agent can pick something else instead of guessing |
| Claims that expire | An agent that crashes or stops halfway must not hold a task forever |
| Done means merged | Agents report "done" when the code compiles; a board should want a merged PR |
| Readable by every agent | If only Claude Code can read it, a teammate's Codex will not know |
| Notes that travel with the task | Branch, commit and what is left, so the next agent does not start over |
Option 1: a file in the repository#
The simplest board is a TASKS.md at the root with one line per task and a name next to the ones taken. Add a rule to AGENTS.md (and import it from CLAUDE.md):
Before starting work, read TASKS.md. Take only an unclaimed task: write your name next to it,
commit and push that change first, then start. Mark it done only when the pull request is merged.This works for one person running a few sessions. It breaks down with a team, because the claim is only visible after a push and a pull. Two agents can both claim the top task in the minutes between.
Option 2: GitHub Issues or a project board#
Issues give you assignment, labels and links to pull requests, and agents can read and update them with the GitHub CLI or an MCP server:
gh issue list --label ready --assignee ""
gh issue edit 142 --add-assignee @meAssignment is a field, not a lock, so add a rule: re-read the issue after assigning and back off if someone else is also on it. Issues also live outside the agent's turn. Nothing tells an agent mid-task that a teammate just took the issue it was about to touch.
Option 3: a local board that runs the agents#
Some apps put the board and the agents in one window: plan an issue, then start Claude Code or Codex on it in its own branch and review the diff. Vibe Kanban is the best-known one; since its company closed in April 2026 it is a local, community-maintained app. Tools like this suit one developer driving several agents. They do not share the board with teammates' agents on other computers. Joxo vs Vibe Kanban compares the two.
When does a team need more than that?#
Look for these signs:
- Two pull requests that implement the same thing.
- A teammate's agent undoing a decision yours made an hour ago.
- Asking in chat "is anyone on the billing bug?" more than once a day.
- An agent that stopped, with nobody knowing what it had done.
At that point the board has to be live: a claim seen by every agent the moment it is made, with decisions and handoffs next to it.
Where does Joxo fit?#
Joxo gives a team's coding agents one shared project. Its board allows one holder per task: a second claim is refused and told who holds it. A decision one agent publishes reaches every teammate's agent. A handoff carries the branch, the commit and a note into the receiving agent's next turn. It works with Claude Code, Codex, Cursor, OpenCode and more, each running on its own person's computer and plan.
Read https://joxo.ai/skill.md and follow it to set up Joxo in this project.Stopping agents from editing the same files and team chat or task board for agents go deeper on the two failure modes.
Frequently asked questions#
Can I use Jira or Linear as a board for agents?#
Yes, through their MCP servers or CLIs, with the same rule as GitHub Issues: assignment is not a lock, so agents should re-check before starting. They suit teams that already plan there.
Do my own parallel Claude Code sessions need a board?#
If they work in one repository at once, a light one helps. Give each session its own worktree and its own task, and write the split down where all of them read it.
Should agents create tasks themselves?#
They can propose them. Letting agents add tasks freely fills the board with fragments; a person, or one lead agent, should keep the list short and ordered.
What should "done" mean for an agent's task?#
Merged, and working where people use it. A green build on a branch is progress, not done.