Parallel agents — worktrees

asd worktree gives each unit of work its own git worktree and a plan/<name> branch. Parallel agents then get isolated files and HEAD — they can’t clobber each other’s working tree — while still sharing project context through the ASD sidecar and (if you use it) the CTXone hub. It mirrors ctx worktree in CTXone.

Lifecycle

asd worktree start <plan>            # add ../<repo>-wt-<plan> on plan/<plan>
# open your agent session in that directory and work there
asd worktree list                    # this repo's plan-scoped worktrees and clones
asd worktree finish <plan> --push    # merge back, push, then tear down

start <plan> creates ../<repo>-wt-<plan> on a fresh plan/<plan> branch and prints its path. New worktrees auto-enable the repo’s .githooks.
list recovers the plan ↔ worktree binding from git worktree list.
finish <plan> merges the plan’s branch back and, by default, tears the worktree down (force-remove + delete branch + prune). --keep merges without teardown; --push pushes the merged branch.

Options

--from <ref> — branch the worktree from a ref other than main.
--shared-target — share one Rust build cache across worktrees by pointing target-dir at <repo>/.wt-target, avoiding a multi-GB target/ per tree.
--clone — isolate via a fresh clone with its own .git (../<repo>-clone-<plan>) instead of a worktree, for remote/cloud agents on another machine; merge-back happens over origin rather than a local merge.

When to use it

Reach for a worktree whenever two or more agents (or you and an agent) work the same repo at once, or when a long-running task shouldn’t block other edits. Each worktree is a clean, isolated checkout — no stashing, no branch-switch churn, no half-applied edits leaking between tasks.