Solutions
Stop parallel agents from undoing each other's work
Collisions have two causes and both need answering. On disk, give every agent its own git worktree, so they are editing different files and physically cannot overwrite each other. On the plan, put the work on a board that leases a task to one agent at a time, so two never pick up the same thing. Forkbench does both, free. The reason it exists, though, is the collision nobody counts: separate worktrees still share one environment, so every agent you started can read the same tokens, and that is the one failure that does not get better when you notice it.
Two agents in the same checkout will eventually overwrite each other, and the failure is quiet: nothing errors, one of them just rewrites a file the other had already fixed, and you find out at review time. It gets worse with a shared plan, where both pick up the same piece of work because neither knows the other started it. Starting several agents is easy. Keeping them off each other is the part nobody warns you about.
Each Thread holds its own terminal tabs, folders, notes, and secrets, scoped so contexts never bleed, and that scope is the point: a key checked into one job is not in another, and a Thread can be locked to its folders by the macOS kernel so an agent cannot read outside them. A live git panel always reflects the branch you're focused on. Every Thread carries a board they pull from: they claim tasks, work in parallel, and hand off, with you as the lead: no agent can verify its own work, so you accept what they report and jump in to unblock anyone. It's agent-agnostic: Claude Code, Codex, Gemini CLI, Aider, and custom scripts all run side by side.
How it works
- 1Group your terminal sessions and agents into Threads by feature or epic.
- 2Isolate risky edits with one-click git worktrees from the visual git panel.
- 3Give the Thread an objective so agents share one backlog and never collide.
- 4Watch the whole board from one window, and reach any agent from the web or your phone.