@mmerterden/multi-agent-pipeline 20.6.0 → 20.8.0

package/pipeline/skills/shared/external/shareplay-activities/evals/evals.json

@@ -3,44 +3,43 @@
   "evals": [
     {
       "id": 1,
-      "prompt": "I'm adding SharePlay to a SwiftUI movie app. Show the Group Activities setup, the activity type and metadata, how to start the activity from the app, how to listen for sessions, and how to sync AVPlayer playback. Keep custom messages only where they are actually needed.",
-      "expected_output": "A source-grounded SharePlay media implementation plan that uses the Group Activities capability, a codable GroupActivity, prepareForActivation/activate or GroupActivitySharingController, long-lived session observation with join/leave cleanup, and AVPlaybackCoordinator for player rate/time sync.",
+      "prompt": "We run a documentary streaming app. Add SharePlay so friends on a FaceTime call can watch a film together, and so someone who is not on a call yet can start one from the film page. Show the setup, the activity and the session code.",
+      "expected_output": "Capability and entitlement setup, a codable watch-together activity, an eligibility-driven start path, a long-lived session listener with join and cleanup, and AVPlaybackCoordinator for transport.",
       "files": [],
       "assertions": [
-        "Adds the Group Activities capability to the app target and references the `com.apple.developer.group-session` entitlement without inventing unsupported Info.plist keys.",
-        "Defines a `GroupActivity` with codable stored data, concise `GroupActivityMetadata`, `.watchTogether`, and a fallback URL or equivalent content identifier.",
-        "Uses `GroupStateObserver.isEligibleForGroupSession` to choose between direct activation and presenting `GroupActivitySharingController` when no conversation is active.",
-        "Observes `Activity.sessions()` in a long-lived manager, stores cancellable `Task<Void, Never>` handles, calls `session.join()`, and cleans up on invalidation or leave.",
-        "Uses `player.playbackCoordinator.coordinateWithSession(session)` for AVPlayer rate/time synchronization and avoids redundant play/pause/seek messenger messages.",
-        "Keeps app-specific `GroupSessionMessenger` messages separate from AVPlayer transport sync."
+        "Adds the Group Activities capability with com.apple.developer.group-session and does not invent Info.plist keys",
+        "Defines a Codable GroupActivity whose metadata uses .watchTogether and a fallback URL",
+        "Uses isEligibleForGroupSession to choose between prepareForActivation/activate and GroupActivitySharingController",
+        "Observes sessions() from a long-lived object holding Task<Void, Never> handles, calls join() and cleans up on invalidation",
+        "Calls coordinateWithSession on the player's playbackCoordinator and does not send redundant play, pause or seek messages",
+        "Keeps any app-specific messages separate from playback transport sync"
       ]
     },
     {
       "id": 2,
-      "prompt": "Review this collaborative whiteboard SharePlay design: every stroke is sent reliably, image drops are sent through GroupSessionMessenger as Data, late joiners rebuild from whatever messages they happen to receive, and the sessions() listener is in a SwiftUI view task. Fix the architecture.",
-      "expected_output": "A correction-focused SharePlay architecture that separates reliable and unreliable message streams, uses GroupSessionJournal for large Transferable attachments, sends state snapshots to late joiners, and moves session observation into a retained manager.",
+      "prompt": "Review our whiteboard SharePlay code: the view's .task iterates Board.sessions(); every pen stroke, cursor move and the Clear button go through one reliable messenger; users paste screenshots, which we send as Data in a message; people who join late see an empty board; and we never call leave.",
+      "expected_output": "A review that moves session observation, splits delivery modes, respects size limits, uses the journal, snapshots state for late joiners and cleans up.",
       "files": [],
       "assertions": [
-        "Moves `sessions()` observation out of a recreated SwiftUI view and into a retained manager or model object.",
-        "Uses unreliable messaging for high-frequency ephemeral stroke or cursor updates and reliable messaging for durable state-changing commands such as clear canvas or selections.",
-        "Mentions that messenger payloads should stay under 256 KB and should not carry large image/file data.",
-        "Uses `GroupSessionJournal` for larger non-time-sensitive attachments, notes `Transferable` item requirements, and includes the 100 MB attachment limit.",
-        "Handles late joiners by detecting new participants and sending a full state snapshot or equivalent canonical state.",
-        "Includes leave/end and task cleanup so abandoned sessions and message listeners do not leak."
+        "Moves sessions() observation into a retained manager rather than a view task",
+        "Sends strokes and cursors with .unreliable delivery and clear or selection changes with .reliable",
+        "States the 256 KB messenger limit and that image data does not belong in messages",
+        "Uses GroupSessionJournal with Transferable items and mentions its 100 MB limit",
+        "Sends a full state snapshot to late joiners",
+        "Adds leave or end handling and cancels tasks during cleanup"
       ]
     },
     {
       "id": 3,
-      "prompt": "I want a SharePlay launch path that works from a normal SwiftUI share button, over AirDrop when people are nearby, and from a custom UIKit button when no FaceTime call is active. Also call out what belongs in GameKit, TabletopKit, or AVKit instead of this SharePlay layer.",
-      "expected_output": "A boundary-aware SharePlay UI plan that uses Transferable for ShareLink/AirDrop/share sheets, GroupActivitySharingController for no active conversation, and keeps Game Center services, TabletopKit board mechanics, and AVKit playback UI or media player details in sibling framework domains.",
+      "prompt": "Our board game app wants three ways to start a SharePlay match: a ShareLink on the lobby, AirDrop to someone nearby, and a button during a FaceTime call. We also need matchmaking, leaderboards and turn rules. What belongs in the SharePlay layer?",
+      "expected_output": "Transferable activation paths, correct choice of controller versus direct activation, discovery-quality metadata, and clear boundaries with GameKit, TabletopKit and AVKit.",
       "files": [],
       "assertions": [
-        "States that SwiftUI `ShareLink`, SharePlay over AirDrop, and system share sheets require the activity or shared item to conform to `Transferable`.",
-        "Keeps the transferable payload minimal by sharing IDs, URLs, or activity descriptors instead of large media or documents.",
-        "Uses `GroupActivitySharingController` when no FaceTime call or Messages conversation is active and direct `prepareForActivation()`/`activate()` when one is active.",
-        "Mentions the `shareplay` SF Symbol or equivalent UI affordance and concise activity metadata as discoverability guidance.",
-        "Keeps GameKit authentication, leaderboards, matchmaking, and voice/chat out of the SharePlay implementation except for handoff boundaries.",
-        "Keeps TabletopKit board equipment/seat/action mechanics and AVKit playback UI details outside the SharePlay layer while allowing GroupActivities session coordination handoffs."
+        "Makes the activity Transferable for ShareLink, AirDrop and share sheets",
+        "Keeps the activity payload minimal, such as an identifier",
+        "Uses GroupActivitySharingController when no conversation is active and direct activation when one is",
+        "Uses the shareplay SF Symbol and concise metadata for the button",
+        "Leaves matchmaking and leaderboards to GameKit and turns or rules to TabletopKit (and playback UI to AVKit), keeping only the hand-off in SharePlay"
       ]
     }
   ]