Guide · 12 min
Organize parallel work without losing control.
Separate independent scopes, use worktrees, request focused commits and integrate every workstream from evidence.
In this article
Objective
Divide a delivery into workstreams that advance together without competing for files, decisions or a shared source of truth.
Before you begin
Map the delivery, affected modules and available checks. Identify shared interfaces first because they must be frozen before dependent work begins.
1. Draw real boundaries
A workstream needs an outcome, exclusive files and an acceptance condition. If two tasks must edit the same central file, they are not independent yet.
Separate investigation from implementation when the diagnosis could change scope. Explicit dependencies are better than apparent parallelism.
2. Use worktrees intentionally
Each agent receives its own branch and worktree. This separates history and files but does not sandbox processes, network, credentials or local services.
Record progress and send dependencies through the app flow. Do not duplicate another workstream just because its result has not arrived yet.
3. Require reviewable commits
The commit should match the declared scope, follow relevant checks and include an objective report. Unrelated changes remain outside the delivery.
Expected result
Workstreams arrive as small units with stable interfaces and enough evidence to make integration a deliberate decision.