Skip to content

Tasks

A local task tracker your agents read and write.

Tasks is a task tracker stored on your computer, in the app's data folder. Open it from the sidebar's Tasks tab, from Tasks at the foot of Agent mode's rail, or with Open Tasks in the command palette.

  • Tasks have an ID (TAN-1, TAN-2…), a title, a description, a status (Backlog, To do, In progress, In review, Done or Canceled), a priority, labels, an assignee and a project, with threaded comments and a history of every change.
  • Quick add puts a task in To do. The list groups tasks by status. Archive a task with its ×, and bring it back from Show archived.
  • Edits from anyone show up at once. When an agent or another window changes a task you're editing, its other fields update and your unsaved text stays. If they changed the same field, Vibezeo says so and you choose Keep mine or Take theirs.

Agents

Agents in your threads and teammate chats read and update the same tasks through Vibezeo's tasks server, and so do Claude Code and Codex in terminal panes. Every change lands in the task's history with who made it, so you can see what an agent did. Threads on an SSH host or in WSL don't get the tasks server: it runs on this computer.

Plan with Lead

Give Lead a goal and it splits it into tasks. On the Tasks page, choose Plan with Lead…, write the goal, and pick a workspace on this computer. Lead plans in a new chat in that workspace, where you can follow its reasoning and answer it. It reads the repository but changes nothing, and it can't write tasks from there. When Lead waits for your answer, the plan says so. After 30 minutes of work, Vibezeo stops Lead and the plan says why.

Lead suggests an engine for each task. It sees which engines are installed and how much of each account's usage window is used. The plan comes back as a list you can edit: change titles and descriptions, move a task to another engine, or remove it. Characters that can hide text are taken out of it. Nothing is created until you choose Create tasks. The tasks go into a new project named after the goal, and each remembers its engine. Their history shows Lead wrote them.

Execute all… first lists what it would start. For each task it shows the engine and the account with the most room left, that the task edits in its own worktree, and which engine reviews the work. A task that isn't To do, or that already has a run, is left alone, and the list says why. Nothing starts until you choose Run. Then each listed task runs in its own worktree, at most 4 at once. Review and merging stay with you: finished work shows in the Dashboard's review queue.

Once its tasks are created, the plan lists each task with its latest run: not started, queued, running, ready to review with the review's verdict (Approved or Changes requested), stopped, failed or done. A line above counts them, and each row opens its task. In the Tasks page's header, a started plan shows as working while its runs go, waiting while one waits on you, and done when they've all ended. When runs are approved, Land N approved… opens Land worktrees with exactly those worktrees chosen, in the plan's order. You check the forecast and confirm the landing there, as with any landing.

A plan stays in the Tasks page's header until you remove it, and Vibezeo keeps it across restarts. If Vibezeo closed while Lead was planning, the plan says so and offers Plan again. Remove plan leaves its tasks, their runs and Lead's thread in place.

Claims

When several agents work in worktrees of one repository, each can say which files it's about to change. An agent claims files or folders with the tasks server's claim tool, saying what it's doing, and checks list_claims for what agents in the other worktrees claimed. A worktree's briefing asks its agent to do this, and so does a task run's first message. Claude Code in a terminal pane claims a file just by editing it, and so does an agent in a chat thread (Claude Code, Codex, or an ACP agent whose edits name their files), so agents that skip the tool are still counted. An edit you refused, or one that failed, claims nothing. Only the file's path is kept, never the change.

When two worktrees claim the same file, or one claims a folder holding a file the other claims, both worktrees' terminal panes show a line like "add-oauth and fix-auth both claim src/auth/session.ts". The Dashboard's review queue lists every overlap above its items. A worktree's claims go when it's merged or removed, and the main checkout's when its last agent session ends. What one agent says it's doing reaches the others marked as untrusted text, never as instructions.

Running tasks

Tracking tasks and running them both work with no account. Execute… on a task runs it for you, in its own lane with progress posted as comments and an optional review, by a different engine when you have one.