Summary
The worktree location convention (.worktrees/<slug>/ inside the repo, never a sibling directory) is documented in J5 Stage 1 and kernel trigger #9, but is not being followed by the facilitator in practice. This is a recurring failure: dogfooding findings F006 and F023 document the facilitator creating worktrees as a sibling of the repo (../maestra-<slug>) instead of inside .worktrees/, across multiple sessions.
Most recently observed again: a fresh session created the worktree outside .worktrees/ despite the documented convention. The current guidance is not emphatic enough to be internalized by the agent.
The goal is to make the convention harder to bypass — whether through stronger wording in the kernel/protocol, a hook that validates the worktree path, or another mechanism that ensures the facilitator declares the worktree in the correct location every time.
Summary
The worktree location convention (
.worktrees/<slug>/inside the repo, never a sibling directory) is documented in J5 Stage 1 and kernel trigger #9, but is not being followed by the facilitator in practice. This is a recurring failure: dogfooding findings F006 and F023 document the facilitator creating worktrees as a sibling of the repo (../maestra-<slug>) instead of inside.worktrees/, across multiple sessions.Most recently observed again: a fresh session created the worktree outside
.worktrees/despite the documented convention. The current guidance is not emphatic enough to be internalized by the agent.The goal is to make the convention harder to bypass — whether through stronger wording in the kernel/protocol, a hook that validates the worktree path, or another mechanism that ensures the facilitator declares the worktree in the correct location every time.