bug(undo-close): a restored parent's live orchestration children keep the dead parent id #1373

Problem

When a parent pane P that has live orchestration children is closed and then restored with Undo Close, the restored pane gets a new session id P′. Its live children keep orchestrationParentId: P.

restoreSessionEntry / restoreTabEntry remap lineage only within the restored entry and the undo stack, never in other sessions' pointers. Replace and Reload Agents do rewrite those pointers, through remapSessionsRelationships. The renderer's visibility gate (orchestrationMcp.ts:306-319) compares ids for equality, so P′'s list_agents / read_agent / send_prompt cannot see or reach its own surviving live children. They also render as top-level rows.

Evidence

• Found by review c of fix(orchestration): closed children follow their parent's replacement #1369 (temp/review-1369/report-c.md, Suspicions). fix(orchestration): closed children follow their parent's replacement #1369 carries main's tombstones of CLOSED children on Undo Close, so the restored parent sees its closed children but not its live ones. That is the live-half analogue of bug(orchestration): reload, switch provider, resume and rewind strip a child's orchestration metadata, so wait_agents reports a false done #879 on the undo path.
• Nothing is worsened by fix(orchestration): closed children follow their parent's replacement #1369: before it, the restored parent saw neither.

Likely fix

Apply remapSessionsRelationships(sessions, successors) at the Undo Close commit, as the replace path does, so live children follow the pane. Main's carry is already wired by #1369.

🤖 Generated with Claude Code