Which change in these updates matters most to ordinary users?
Most directly, the October 8 approval behavior: when a Scheduled Task reaches a tool the organization marks "ask," it used to be skipped, and now the run shows an approval prompt and waits. A step that used to be silently passed over can now leave a run standing still. Whether it affects you depends on whether your organization marks tools as ask.
Why would Anthropic change "skip" to "wait"?
The changelog does not give a reason. A reasonable inference is that a tool marked ask means the organization wants a person to confirm before it runs, and silently skipping it lets a task do less without anyone knowing, while waiting preserves the intent of human review. This is inference, not an official statement.
Why does a newer run retire an earlier run's prompt?
Per the changelog, if an earlier run of the same task left an approval prompt nobody answered, the next run retires that old prompt. The effect is that a task does not accumulate a pile of stale prompts. The changelog does not say whether the older run itself is cancelled or keeps waiting, so do not assume either.
What should I do now?
Update to v2.42304.0 or later. List your Scheduled Tasks and find those that use tools marked ask. Check whether any same-named task had its prompt overwritten before October 8. If your computer has had workspace start problems, watch for the "Scheduled Task failed" notification, since tasks may be running without the workspace.
Between October 8 and 15, the Claude Desktop app shipped three updates (v2.31226.0, v2.37337.1 and v2.42304.0), and a large share of the Cowork fixes in the official changelog point at one place: Scheduled Tasks running when nobody is there.
v2.31226.0 fixed four things. First, when a Scheduled Task reached a tool that the organization's approval policy marks "ask," it used to be skipped; now the run shows the approval prompt and waits, and a newer run of the same task retires an earlier run's unanswered prompt. Second, Claude re-requesting access to a folder the session already had, which left scheduled runs waiting on an approval nobody was there to give, is fixed. Third, creating a scheduled task while the task list was still loading could overwrite the prompt of an existing task with the same name; that no longer happens. Fourth, a "Scheduled task failed" notification now appears when a task's Sandbox cannot start.
v2.37337.1 fixed local scheduled tasks that could not be edited after their first run, which showed "Failed to update scheduled task." Another fix covers computers where Claude's workspace fails to start: due scheduled tasks now start without the workspace after about ten minutes, and commands that need the workspace fail in those runs. The same release also fixed scheduled tasks failing to load when the Windows profile lives in an FSLogix-style container.
v2.42304.0 improved the import from the third-party app: copied scheduled tasks keep their attached files, and where device management allows it, their folders and, while a local MCP Server is unchanged, the approvals you gave for its tools.
The changelog describes behavior changes only. It does not explain how the "ask" policy interacts with the October 1 default of automatic approval for scheduled tasks Claude creates, and it does not say whether a waiting prompt ever times out. Do not assume either.
If you rely on scheduled tasks, first confirm the desktop app is on v2.42304.0 or later. Then check three things: which tasks use tools your organization marks "ask," since those runs may now sit overnight waiting for approval; whether any same-named task had its prompt overwritten before October 8; and, if your computer ever failed to start the workspace, whether scheduled runs are now executing without it and failing, which the failure notification should show.