A routine's Run now says what it will spend, before it spends it (fix #1506) - #1636
Conversation
Hovering a Run now on the Routine work card now names the three things the card could not otherwise show: what that routine does, which model its start would use, and where it runs. All three come from the Global options a page away, so the button's cost was invisible until the agent already existed. The first line is the preset's own sentence rather than one written again for this card, carried onto the job like its label already is, so the launcher and the routines list describe the same work with the same words. Two presets had no such sentence, and now do. The drain row answers differently, because its Run now is the sweep: it visits every project the daemon watches, and each of those resolves its own committed settings on top of these, so the model and place the other rows promise would both be a guess there. The model lists, the run-target names and the "nothing pinned" wording move to `dashboard/lib/agent-settings.ts`: four surfaces name them now, and a tooltip that reports settings has to say exactly what the launcher says.
The launcher's preset hints rendered with the right words and were covered by the very menu that triggered them, so the one-liner a preset carries could not be read from the place it is offered. `z-50` sat on the popup, which Base UI leaves `position: static` — where a z-index does nothing at all. The menu and the popover both carry theirs on their own positioner, so both won. It moves to the positioner here too. Found by hovering the two preset one-liners this branch adds; it is not new, and it was true of every tooltip opened over a menu.
|
🤖 automated A second commit: the launcher's preset hints were being painted behind the menu that triggered them. Not a regression from this branch — I found it by hovering the two preset one-liners the branch adds, and it was equally true of the ones that shipped months ago.
Measured in the page before changing anything, with the "Suggest new features" hint open inside the presets menu:
So the hint was rendering correctly and losing the paint. Setting the z-index on the positioner in the live page flipped that same probe to the tooltip itself, which is exactly what the one-line change does. Fixed, rebuilt, and re-checked in the browser: No test. jsdom computes no stacking and the classes carry no CSS there, so the only assertable thing would be the class name — which would pass whether or not the tooltip is visible to a human, and would have passed for the whole time this was broken. The reason it must sit on the positioner is written where the change is instead. |
…1506) (#1636) * A routine's Run now says what it will spend, before it spends it (#1506) Hovering a Run now on the Routine work card now names the three things the card could not otherwise show: what that routine does, which model its start would use, and where it runs. All three come from the Global options a page away, so the button's cost was invisible until the agent already existed. The first line is the preset's own sentence rather than one written again for this card, carried onto the job like its label already is, so the launcher and the routines list describe the same work with the same words. Two presets had no such sentence, and now do. The drain row answers differently, because its Run now is the sweep: it visits every project the daemon watches, and each of those resolves its own committed settings on top of these, so the model and place the other rows promise would both be a guess there. The model lists, the run-target names and the "nothing pinned" wording move to `dashboard/lib/agent-settings.ts`: four surfaces name them now, and a tooltip that reports settings has to say exactly what the launcher says. * A tooltip is drawn above what it opens over (#1506) The launcher's preset hints rendered with the right words and were covered by the very menu that triggered them, so the one-liner a preset carries could not be read from the place it is offered. `z-50` sat on the popup, which Base UI leaves `position: static` — where a z-index does nothing at all. The menu and the popover both carry theirs on their own positioner, so both won. It moves to the positioner here too. Found by hovering the two preset one-liners this branch adds; it is not new, and it was true of every tooltip opened over a menu.
1 participant
Closes #1506.
The problem in one line:
Run nowon the Routine work card spends an agent on a model and a place that are nowhere near the button. #1634 gave you a way to go and set them; this says what they currently are, before you press anything.What it looks like
Hovering any
Run now:Three lines for the issue's three questions — what happens, which model, where it runs — plus the one fact that is arguably the biggest hidden cost: a card-fired routine runs unattended (#1279), so nothing will be asked mid-run.
Two decisions worth your eye
The first line is the preset's own sentence, not new copy. Presets already carry a one-liner that the launcher's menu shows on hover; the routine now carries it too, exactly the way it already carries the preset's label. So the launcher and the card describe the same work in the same words, and rewording a preset rewords both.
Two presets had no such sentence and now do — Plan tickets ("Turn
tickets/*.mdinto costed plans (tickets/*.plan.md)") and Drain queue ("Work the entries already on the queue (TODO_AGENTS.md)"). Plan tickets is in the launcher menu, so that one is user-visible there too. Reword either if you want them said differently.The drain row reports something else entirely, and it has to. Its
Run nowis the sweep, not a start, and two things the other rows promise are simply untrue of it:the-framework.ymlon top of these preferences (resolveProjectAgentOptions,daemon-services.ts:149), so naming a model here would be a guess.So it says:
The number is the live concurrency setting from the same card.
How it's built
The model lists, the run-target names and the "nothing pinned" wording moved into
dashboard/lib/agent-settings.ts. They were spread across four surfaces — the launcher's driver tree, the gear's "Run on" list, the Settings page and now this card — and a tooltip that reports settings is only worth anything if it says exactly what the launcher says. One side effect: the Settings page called the local machine "This device" while the gear called it "This machine"; there is one name now.A model is still only named inside its own driver's list (#1143): nothing pinned, or a model pinned on the other driver, reads as "the CLI's own default" rather than borrowing the first entry.
Verification
Dogfooded on the real daemon, both rows and both refactored surfaces:
Run nowon "Update from GitHub" → the three lines above, withClaude Code · Opus · This machinematching the actual preferencesRun now→ the sweep wording instead, with the live concurrency of 2Tests: 4 new, each proven by breaking the code it guards. Dropping the preset's sentence, falling back to the first model when none is pinned, hardcoding the run target to local, giving the drain row the ordinary settings line, and ignoring the concurrency in its fan-out line each failed exactly the intended test and nothing else. Full suite green: 1536 node + 792 dashboard.
RoutineWork.SPEC.md, its test SPEC,preset-catalog.SPEC.md, the newagent-settings.SPEC.mdandFEATURES-SPEC.mdall updated.