Repository navigation
feat(tars): read a work card in three lines - #1018
Merged
Merged
Conversation
astyfx
requested review from
heath-s and
png-1105
and removed request for
a team
November 19, 2026 07:52
Upwind code scan | ✅ Passed
View full scan details in Upwind Console → Scan history (3 commits)
Last updated: |
astyfx
force-pushed
the
feat/task-panel-single-view
branch
from
November 19, 2026 12:08
ec62355 to
b18da9e
Compare
astyfx
force-pushed
the
feat/tars-card-three-lines
branch
from
November 19, 2026 12:08
3efb267 to
0af6fa9
Compare
A Work tray card now reads in three lines: the work's title, one sentence
(what it does now, what it needs from the user, or what came out), then
its state with exactly one action. The state is the task's five user
states: card-view.ts translates Tars's own states and Fleet's rows into
resolveTaskUserState's facts, so the card has no state words of its own.
The action fits the state: Confirm or Answer while it waits on the user,
Open report for work Tars could not finish, Resume or Hand back for held
work, Start now for queued work, Open task for running work, and the
pull request or report for finished work. Answer shows the work's
decision card above the message box.
The title is a disclosure for the card's detail, which keeps everything
that left the card: owner, turns left, helpers, usage, branch, model,
the stage with its clock and strip, the goal, the observed line, recent
progress, what waits on the user, Follow up, Create draft PR, the
state's own control, Open task, the more-actions menu (take over, queue
moves, Show in Fleet, Cancel work), the step count, Request and
Timeline. A queued card keeps its reorder grip on the state line.
The same work's progress is no longer repeated on the screen. The
conversation's flow block is the delegation's record, folded to the
work's title without a stage count, state or live line. The composer
keeps one status line ("2 running now", with the quiet count), which
opens the Work tray on the running card instead of unfolding a second
list of the work.
Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The base gained the on-hold user state. A card whose run waits for a usage reset with a known time, or for the outside event it asked to be woken by, now reads Waiting with the run's own sentence saying when it resumes, and offers no action: nobody has to act. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… card A card whose work waits on a helper showed Waiting for you, but its Open task opened the work's own task. The action now carries the request's own task and workspace and goes through the view's request path; the detail keeps Open task for the work's own task. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
A pull request ready to merge made running work read Waiting for you, and a turn that went quiet while still active made it read Blocked. Only a real request in the task (question, approval, sign-off) and a real stop (Tars block, failed turn, blocked or stuck stage) feed the shared state now; review rows stay in the card's detail. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…tion While a result waited for the user's check, every Tars decision on the work was dropped with the result-review row Stave adds, so a real question read as a sign-off with Open report. Only that added row is left to the state now (`tarsStateSignalId`); a question's Answer and a proposal's Confirm come first, and the added row never takes a proposal's Confirm. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… prepared Preparation saves the task id before it prepares the workspace and creates the task, so the card's one action opened a task that did not exist yet and failed. Until the work's run exists the card shows no action and its detail drops Open task; Pause stays in the detail. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…shed The card decides its state on every render, and it re-renders with each step its work streams. The facts walked every stage record and changed path for a count only finished work reads; running, waiting and blocked work no longer read the run's stage records for it. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
With the tray grouped by project and a group folded, the status line above the composer (and a start-screen count) moved focus to a card inside the hidden group, so focus stayed where it was. The group now opens with the lane, then the card's title takes focus. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…card e2e `cardStates=true` adds work on hold, work waiting on a helper's approval, running work under review-only Fleet rows, a question about a ready result and work still being prepared. The e2e checks each reads in three lines with the right state and action, that a helper's request opens the helper's task, and that the status line reaches a card in a folded project group. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…work card's state Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… widths The card is a grid and the slot of its second line had no minimum of its own, so a live line too long for the card (the stale-progress note in Korean at 390 px, in both languages at 360 px) widened the whole card past its lane and pushed the title's chevron and the action out of view. The slot now gives up that minimum and the line ends in an ellipsis. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…its for A card for work still being prepared read "Working" over "The work has not started yet", and queued work read "Working" too. Neither runs and neither needs the user, so both are on hold now: queued work says the scheduler's reason and keeps Start now, work being prepared says it waits for its workspace and offers nothing. A request in its task or a stop still wins. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
Taking work over is the user's own choice, so its card no longer wears the warning mark and "Blocked". It is on hold in the neutral waiting mark, says that the user holds it, and keeps Hand back as its action. What Fleet lists for its task meanwhile leaves the card as it is, as it did before; a decision Tars asks still waits on the user. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The Needs you header, the tray's indicator and the sidebar's Tars entry counted Fleet's rows, so work with two rows (a question and its result review, or two review rows for one task) made the header say more than the cards under it: "Needs you 6" over four cards. Every lane now counts its cards; a card's detail still lists each of its rows. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
The detail check read the chevron's box right after the click that opens the card. While it turns its box is wider than the icon, so the check failed by under two pixels whenever the measurement landed mid-turn. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
…ard stays in the tray Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
A pause is a choice too, so its card no longer reads "Blocked" under the warning mark: it waits in the neutral state, says that it is paused (or the words saved with the pause) and keeps Resume. A pause that ended with Tars no longer in charge of the task carries its reason and stays blocked. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
… work in its task A work card decided its sentence before Tars learned to observe a run the user starts in a delegated task. Such work now keeps the state its run has and says whose run it is: Working with the observing sentence instead of the live line, Finished with "Tars did not verify it" ahead of an earlier pull request, and Blocked leading to the task when the user's run ended short of the goal. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com>
astyfx
force-pushed
the
feat/tars-card-three-lines
branch
from
November 19, 2026 12:25
0af6fa9 to
0adfd02
Compare
astyfx
enabled auto-merge (squash)
November 19, 2026 12:25
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.

Summary
Stacked on the Task panel change (base
feat/task-panel-single-view). A Tars work card showed about 24 lines and 6 buttons, and the same progress appeared in four places on one screen (synthesis #9).resolveTaskUserState; queued, preparing, paused and taken-over work read as on hold in neutral tone.Review and verification
Written and reviewed by different providers: each branch was implemented by one model, reviewed independently by a model of the other provider, and fixed until the reviewer's blocking findings were resolved. Lead checks on the rebased head:
bun run typecheck,bun run lint,bun run check:i18n,bun run check:doc-paths,bun run check:max-lines-ratchet,bun run build:pages,git diff --check, and every changed unit test file run one per process: all exit 0. New regression tests fail on the previous code and pass now. Not verified: the installed Electron app, live providers, real macOS sleep, real GitHub calls.Part of the design-review follow-up (synthesis priorities 1–14).
🤖 Generated with Claude Code