Agent Team recovery cannot retain an existing dedicated worktree

Summary A replacement Agent Team writer cannot safely resume a dirty dedicated worktree after the original writer is closed.

Reproduction (1DevTool 1.70.0, macOS) 1. Launch a writer with run --terminal --category=implement; it receives a dedicated worktree. 2. Interrupt or transfer the writer while its Task Card has uncommitted, allowlisted changes. 3. Attempt to launch a replacement for the same worktree using run --terminal --cwd=.

Actual result The terminal-team start payload does not forward cwd or worktreePath. A non-write category opens the project root; a write category provisions a new worktree. team start --manifest-stdin transports a manifest but the controller does not apply member.worktreePath.

Backup Agents do not provide a safe fallback: they are Hierarchy/UI-driven, failover requires a human gesture, and fallback terminal creation does not receive the previous worktree path.

A safe external orchestrator must therefore fail closed; spawning first and cleaning up is unsafe because it opens a writer in the primary checkout.

Expected result A replacement or backup seat can be created directly in the exact existing dedicated worktree, preserving the transcript, then claim the prepared lease.

Requested API - Accept an existing worktreePath on Team run --terminal / team start members and apply it when creating the terminal. - Support a controller-owned backup/failover seat with the same exact worktree path and transcript, callable headlessly for a Team seat. - Expose a capability query so integrations can fail closed before any terminal is created.

This blocks reliable recovery of long-running, allowlist-scoped coding work. No secrets, production data, or user files are involved.