Expose an optional placeId on operational tools for deterministic multi-place routing

Back to board
### What problem does this solve? When two or more Studio instances are connected at once — e.g. a separate **lobby** place and the **main world** place open side by side — the operational tools (`manage_terrain`, `manage_ui`, `mutate_instances`, `manage_lighting`, `manage_scripts`, `system_info`→`place_info`, etc.) give no way to choose *which* connected place a command runs on. The dashboard correctly lists every connected plugin, and the connection log shows `Multiple plugin clients connected, routing to most recent`, so commands silently route to whichever plugin was most-recently active. That has two practical failure modes: - **Non-deterministic:** with both plugins idle and sending heartbeats, "most recent" can flip between them, so the same command can land on the wrong place from one run to the next. - **Unreachable for unattended agents:** a headless agent (Claude Code, Codex, Gemini, etc.) has no way to "focus a Studio window," so it cannot influence which place is most-recent. There is no per-call lever to pin the target. Today the only tool that accepts a `placeId` is `manage_sync`, so multi-place targeting exists for the sync layer but not for live operations. ### Proposed solution Add an optional `placeId` parameter to the operational tools' input schemas — the same field `manage_sync` already uses — and pass it through so it can be matched against the connected plugins. When omitted, keep the current most-recent behavior, so it's fully backward compatible. Since routing already falls back from a target to "most recent", this looks like surfacing an existing capability rather than new plumbing. Would round it out nicely: - A `system_info` action (or an extension of `connection`) that returns connected places with their `placeId` / `placeName` / `clientId`, so an agent can discover valid targets before sending a command. - Optionally accept `clientId` as an alternative selector, for when two Studios share the same `placeId` / `dataModelName`. - Have tool responses echo which `placeId` actually handled the call, so agents can verify routing. ### Which tier would this apply to? Pro ### Additional context I'm a Pro user. Scenarios where deterministic per-place targeting matters: - **Lobby + main world** (the standard hub-and-spoke Roblox layout) open at the same time while an agent edits terrain / UI / instances on one specific place. - **Multi-place games** with separate places per level, dungeon, minigame, or biome, where the agent targets a chosen place. - **Parity / template propagation** — applying the same UI, lighting, or module setup across several places to keep them consistent, looping over places by id. - **A/B or variant places** — iterating on two variants side by side and needing edits to land on exactly one. - **Automated CI / playtest harnesses** running across multiple places in a single session, where results must map back to the right place. - **Unattended or parallel agents** — an autonomous agent (or several, one per place) that can't rely on window focus and needs a deterministic selector. Thanks for WEPPY — the multi-Studio connection handling and the dashboard are already great; this would make it fully usable for automated multi-place workflows.