Repository navigation
Replies: 1 comment
|
A case for this that doesn't need a threshold at all: a task bound to a thread whose account is already out of usage. I had a 10-minute interval task posting into a thread while that thread's Claude account sat at its weekly limit. On 11-10, between 02:25 and 06:15 UTC, every tick queued its prompt and failed within about two seconds: 24 runs in a row. Each one added four items to the thread: the task's prompt, the notice "Claude usage limit reached. This turn is paused until the 7-day limit resets in 31h 15m.", Claude's own limit message, and a "Usage limit reached" error. That's about a hundred items of noise in four hours, mixed into the thread's real history. T3 knew the limit and when it resets: it printed the reset time in that notice on every tick. The scheduler doesn't look at it. A thread-bound run queues its prompt with So on top of the opt-in threshold, I'd ask for a default with no setting: when the account a run would use is out of usage, skip the run and log it as skipped with the reason and the reset time, or hold it until the reset. For a thread-bound task that's the thread's current provider, since that's what the run uses. On your open questions: I agree unknown usage should run. For one-time tasks I'd hold until the reset rather than skip, because a skipped one-time task never fires again. Written up with Claude Code from a session on the affected machine. |
Uh oh!
There was an error while loading. Please reload this page.
Note
🤖 Claude Opus 5.5 writing on behalf of XelaJr
Before submitting
Area
apps/server (scheduled tasks), contracts, the scheduled task editor on web and mobile, and the
schedule_taskMCP tool.Problem or use case
I think it would be useful to decide whether a scheduled task runs based on how much usage I have left on the model it needs.
Scheduled tasks spend the same subscription as my own threads. If a task fires when I have 25% left of my Claude 5-hour session, or I'm close to the weekly limit, it uses up what I wanted for interactive work, or it hits the limit halfway through and fails. Today the only way to avoid that is to pause the task by hand and remember to turn it back on.
T3 Code already has the numbers. Every provider reports its windows (
session,weekly,monthly) withusedPercentunder Usage → Limits. The scheduler just doesn't look at them.A gate command from #33242 doesn't fit this well, because a script has no simple way to read T3's usage limits for the account the task runs on.
Proposed solution
An optional usage condition on interval, fixed-time and one-time tasks:
nextRunAtstays unchanged.Open questions
Scope
An optional field in contracts (existing tasks decode unchanged), the check in
ScheduledTaskServicebefore dispatch, a control in the web and mobile task editors, the matching field on theschedule_taskMCP tool, and the scheduled tasks section of the user docs.I'm happy to implement this if you agree with the direction and scope.
All reactions