Worktrees for parallel lanes are created one directory above the repository (../abcd-wt-<lane>, ../abcd-f185, ../abcd-integration), which litters the parent directory with a dozen sibling checkouts named by convention only, leaves them behind when a session dies (21 spent worktrees, 1.4G, were removed by hand on 2026-09-01), and gives nothing a stable place to look for them. Assessment of where they should go: (1) inside the repository under .abcd/.work.local/worktrees/ keeps them with their owner and gitignored, but a worktree inside the main working tree is walked by every tree scan (the payload gate, name-guard, docs-lint), which already flakes during worktree creation (iss-2608261331317889) and would slow every gate; (2) under the user-level home, ~/.abcd/worktrees/<root-sha>/<branch-slug>/, keyed the way the history store already keys a repository by its root commit, keeps them out of every scan, survives a session, and gives an abcd verb a place to list, prune and reattach them; (3) the git default (anywhere) is the status quo. Option 2 fits the existing home layout and the one-canonical-primitive rule (the root-sha keying exists), and it makes the concurrent-sessions convention in AGENTS.md checkable: a session lists its peers' worktrees by reading one directory. It needs: a verb or documented one-liner (abcd worktree add <lane> / list / prune), the AGENTS.md convention updated, the lane brief updated, and a prune rule for worktrees whose branch is merged.

Worktrees for parallel lanes are created one directory above the repository (../abcd-wt-<lane>, ../abcd-f185, ../abcd-integration), which litters the parent directory with a dozen sibling checkouts named by convention only, leaves them behind when a session dies (21 spent worktrees, 1.4G, were removed by hand on 2026-09-01), and gives nothing a stable place to look for them. Assessment of where they should go: (1) inside the repository under .abcd/.work.local/worktrees/ keeps them with their owner and gitignored, but a worktree inside the main working tree is walked by every tree scan (the payload gate, name-guard, docs-lint), which already flakes during worktree creation (iss-2608261331317889) and would slow every gate; (2) under the user-level home, ~/.abcd/worktrees/<root-sha>/<branch-slug>/, keyed the way the history store already keys a repository by its root commit, keeps them out of every scan, survives a session, and gives an abcd verb a place to list, prune and reattach them; (3) the git default (anywhere) is the status quo. Option 2 fits the existing home layout and the one-canonical-primitive rule (the root-sha keying exists), and it makes the concurrent-sessions convention in AGENTS.md checkable: a session lists its peers' worktrees by reading one directory. It needs: a verb or documented one-liner (abcd worktree add <lane> / list / prune), the AGENTS.md convention updated, the lane brief updated, and a prune rule for worktrees whose branch is merged.

Corroboration (2026-09-20, Gropius session gropiusllm-2b, relayed to abcd-17). A third placement, and the one the harness itself defaults to: a managed repository's sub-agent lanes run in worktrees under .claude/worktrees/ INSIDE the checkout, one per intent or issue. That is the shape option (1) above rejects (every tree scan walks it, and worktree creation already races the payload gate, iss-2608261331317889), and it is not the session's choice: the harness's worktree isolation creates them there unless told otherwise, so adr-2609091248200336 and the machine-scoped store are unknown to a managed repository until abcd's own worktree verb (itd-148, spc-42) exists and the lane brief names the path. Until then the convention in abcd's own AGENTS.md (git worktree add ~/.abcd/worktrees/<root-sha>/<name>) is the documented one-liner, and it reaches no managed repository.

Grounds

• pursued: parallel-lane worktrees have one ruled home outside the checkout; shown wrong if a lane worktree is created beside or inside a checkout under the current convention