manage_studio play-state tracking doesn't detect sessions started via Studio's "Test" / "Test Here" / other non-legacy launch modes
Back to board### What problem does this solve?
manage_studio's play-state tracking (play_status, play_start, and by extension manage_logs' live log stream) appears to only detect playtest sessions launched through whatever internal hook it's built against — seemingly the classic "Play" (F5) API. When a session is instead started via Studio's newer ribbon options — "Test", "Test Here", "Run", "Team Test", "Server & Client" — play_status continues to report state: "edit" indefinitely, and no new entries appear in manage_logs, even minutes into an active, visually-confirmed playtest.
This happens despite the plugin's own connection remaining healthy: system_info → connection shows "connectionState": "connected" throughout, with lastSeen/lastRoutedAt timestamps advancing normally. So it's not a dropped connection — it's specifically a missed play-state signal for these launch paths.
Net effect: an AI agent (or any external tool) driving Studio through this MCP has no way to know a playtest is running, inject test scripts into it, or read its console output, whenever the user starts testing through any option other than the one currently wired up. Given some Studio installs/versions apparently no longer surface a distinct classic "Play" button at all (only Test/Test Here/Run/Team Test/Server & Client), this can make automated verification impossible for those users entirely.
### Proposed solution
Extend the playtest-detection hook to fire for all of Studio's playtest launch commands ("Test", "Test Here", "Run", "Team Test", "Server & Client"), not just the one currently supported. At minimum, play_status should transition out of "edit" and manage_logs should start receiving output the moment any of these modes goes active — matching what's already visible to a human watching the Studio window. If different modes need different handling (e.g., "Run" has no client/character, unlike "Test"), the state payload could include which mode is active so callers can adjust expectations accordingly.
### Which tier would this apply to?
Pro
### Additional context
- MCP server v2.10.0, Studio version 0.728.0.7280895, Windows.
- Reproduced repeatedly: play_status never left "edit" across multiple attempts using "Test Here" and then plain "Test", despite the user confirming the character had spawned and the session was interactive.
- The user's Studio ribbon only exposed "Test", "Test Here", "Run", "Team Test", "Server & Client" — no separate legacy "Play" button — suggesting this is tied to a newer Studio UI/version where the old hook no longer applies.
- Practical impact on this session: blocked an AI-assisted debugging workflow. The user needed to verify whether a key-press-triggered raycast was firing correctly in a live session; with no visibility into the running playtest, we had to fall back to manual human-reported testing instead of automated verification via execute_luau.