Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

@xgopilot 解决你提到的问题

Result

Fixed the race I called out in review by changing shell-ball runtime observation handling from per-task late subscription to a global runtime subscription with task-scoped buffering and replay.

What Changed

• Added subscribeAllTaskRuntime(...) in apps/desktop/src/rpc/subscriptions.ts and kept subscribeTaskRuntime(...) as the filtered wrapper for task-detail consumers.
• Updated apps/desktop/src/features/shell-ball/useShellBallCoordinator.ts to queue task.steered / loop.* notifications that arrive before the submit response registers the formal task_id, then replay them once the task is bound locally.
• Extended apps/desktop/src/features/shell-ball/shellBall.contract.test.ts with a runtime-race contract case and updated subscription stubs in apps/desktop/src/features/shell-ball/test-stubs/rpcSubscriptions.ts.

Validation

• Passed: git diff --check
• Not run: pnpm --dir apps/desktop typecheck
• Not run: pnpm --dir apps/desktop lint
• Not run: shell-ball contract test command from the PR description

The desktop workspace in this runner does not have pnpm/corepack or installed node_modules, and .cache/shell-ball-tests/... had not been generated, so I could not execute the requested local checks here.

Git

• Pushed to the PR branch: gdemonc:issue-327-shell-ball-runtime-subscriptions
• Commit: 3d36dbe (fix(shell-ball): buffer early runtime notifications)
• Fork PR helper result: fix(shell-ball): buffer early runtime notifications gdemonc/CialloClaw#8

View job run

@xgopilot 还有什么问题

Review

No additional issues found in the updated patch.

I re-checked the change against origin/main...HEAD, including the new subscribeAllTaskRuntime(...) path in apps/desktop/src/rpc/subscriptions.ts and the task-scoped runtime queue/replay flow in apps/desktop/src/features/shell-ball/useShellBallCoordinator.ts. The earlier race around runtime notifications arriving before local task registration is addressed by the global subscription plus per-task_id buffering, and I did not find another behavioral regression in the changed code.

I did not rerun the desktop checks in this runner.

View job run