fo-dispatch-recovery

# First Officer Dispatch Recovery (Claude) The two Claude dispatch exception bodies, each read only at its own failure trigger in `claude-fo-dispatch.md` — never at boot, never on a session where dispatch never fails. ## Break-Glass Manual Dispatch The resident trigger line already covers the first action (report the helper failure — command, exit code, stderr — to the captain before proceeding). The dispatch mode selected before `dispatch build` remains authoritative: do not probe another transport, retry in the other mode, or turn a selected bare dispatch into a named worker. Populate `{numbered checklist}` with the output of `«dispatch.checklist»(entity, stage)`; do not rebuild its rules here. In either arm, include `model="{effective_model}"` only when an effective model is set. For selected bare mode, use this blocking call. Omit `name`, `team_name`, and `run_in_background` entirely; Claude's observable tool stream may preserve that omission or normalize it to `run_in_background=false`, and both mean blocking bare dispatch. Never pass or accept `run_in_background=true`. Omit the completion-message block because the blocking return is the completion signal: ``` Agent( subagent_type="{dispatch_agent_id}", description="{entity title}: {stage}", model="{effective_model}", prompt="## First action\n\nBefore anything else, invoke your operating contract:\n\n Skill(skill=\"spacedock:ensign\")\n\nThis loads the shared ensign discipline (stage-report format,...