Repository navigation
feat(autonudge): pause and play icon controls in the goal popover - #16030
Conversation
af1f2dc to
ebc813e
Compare
GPT 5.6 Review — ✅ human override acceptedHuman judgment by @pepmach overrides the GPT 5.6 finding for [GPT-OVERRIDE] 3d6c7e2 This comment is updated in place on each push. The model was not re-run because an authorized human decision supersedes it. False positive or not applicable? A repository writer can comment: |
Design Review (Fable 5) — ✅ PASSDesign-level review of Design-Verdict: PASS Verify the product-owner overrides the RFC defers to the PR itself: the Clear-link row placement lane reading, and the First-Principles draft-status lane. Suggestions
[DESIGN-REVIEWED] 3d6c7e2 Description read[DESCRIPTION-READ] 60b7d6fac27040c308987b7e5f7625c94738d26b3260dea9a1e842fb5951e7b5 That sha256 names everything this verdict read: the PR description, plus the 1 evidence file(s) the workflow collected from it and handed this review. It covers that evidence because the media strip replaces every attachment URL with the same placeholder, so an edit swapping one attachment for another would leave the description's captured text identical while what this review judged changed. Recompute it with |
First Principles Review (Fable 5) — ✅ human override accepted@pepmach overrode this lane for |
UX Review (Fable 5) — ✅ human override accepted@pepmach overrode this lane for |
1b573a2 to
b7f4bb6
Compare
Opus 5 Review — ✅ human override acceptedReviewed Human judgment by @pepmach overrides the Opus 5 finding for [OPUS-OVERRIDE] 3d6c7e2 Verdict recorded from an authorized human decision for commit False positive or not applicable? A repository writer can comment: |
b7f4bb6 to
d8513ea
Compare
|
/ai-review override gpt d8513ea: Clear stopped goal is a text link, not a button; the row holds exactly two buttons. The two-buttons-per-row rule guards against button clutter, which this layout keeps. PR owner's layout decision. |
|
/ai-review override first-principles d8513ea: Product shape decided by the PR owner: Pause and Lightning icon controls while running, Play while paused, Clear as the only delete, resume resets the loop budget, limit stops read Paused. The RFC recording it ships in this PR at docs/request-for-change/rfc-goal-popover-pause-play-controls.md. |
Human judgment recorded@pepmach marked the gpt AI finding as false positive, not applicable, or explicitly accepted for
This decision applies only to this commit. A new push requires a new judgment. |
Human judgment recorded@pepmach marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for
This decision applies only to this commit. A new push requires a new judgment. |
d8513ea to
35284f5
Compare
|
/ai-review override gpt 35284f5: Clear stopped goal is a text link, not a button; the row holds exactly two buttons. The two-buttons-per-row rule guards against button clutter, which this layout keeps. PR owner's layout decision. |
|
/ai-review override first-principles 35284f5: Product shape decided by the PR owner: Pause and Lightning icon controls while running, Play while paused, Clear as the only delete, resume resets the loop budget, limit stops read Paused. The RFC recording it ships in this PR at docs/request-for-change/rfc-goal-popover-pause-play-controls.md. |
While a loop exists the popover offers exactly two icon-only controls in every state, Pause and Play, each named in aria-label and title. On a running loop Play saves the pending field edits and fires a nudge now; on a paused loop Pause is disabled and Play saves the edits, resumes the loop on a fresh budget and fires -- one request path, nothing to raise first. A pending edit marks Play with a dot. Clear stays the only way to delete a goal, as a text action on a paused loop behind its existing confirm. The loop's state is one plain-English line by the title and independent of the buttons: the countdown and cycle readout while running, and "Paused · <why>" otherwise, worded per stopped_reason -- you paused it, the agent stopped it, the cycle or time limit was reached (resuming starts a fresh budget), approval is awaited, or the last cycle could not run. The word Stopped, the Save and Stop buttons, the Nudge-now button and every bound help line that named a tool or a config key are gone, with their catalog keys in all twelve locales and the pseudolocale. Tests are consolidated into table-driven cases per state and press, keeping every behaviour assertion; coverage of the two components is at or above the previous head's.
35284f5 to
3d6c7e2
Compare
|
/ai-review override gpt 3d6c7e2: Clear stopped goal is a text link, not a button; the row holds exactly two buttons. The two-buttons-per-row rule guards against button clutter, which this layout keeps. PR owner's layout decision. |
|
/ai-review override first-principles 3d6c7e2: Product shape decided by the PR owner: Pause and Lightning icon controls while running, Play while paused, Clear as the only delete, resume resets the loop budget, limit stops read Paused. The RFC recording it ships in this PR at docs/request-for-change/rfc-goal-popover-pause-play-controls.md. |
|
/ai-review override ux 3d6c7e2: Icon-only controls are the PR owner's design choice; the lightning button is named "Nudge now" / "Save edits and nudge now" on hover and for screen readers, and the dot marks unsaved edits. Visible labels are not wanted here. |
Human judgment recorded@pepmach marked the gpt AI finding as false positive, not applicable, or explicitly accepted for
This decision applies only to this commit. A new push requires a new judgment. |
Human judgment recorded@pepmach marked the ux AI finding as false positive, not applicable, or explicitly accepted for
This decision applies only to this commit. A new push requires a new judgment. |
|
/ai-review override first-principles 3d6c7e2: Product shape decided by the PR owner: Pause and Lightning icon controls while running, Play while paused, Clear as the only delete, resume resets the loop budget, limit stops read Paused. The RFC recording it ships in this PR at docs/request-for-change/rfc-goal-popover-pause-play-controls.md. |
Human judgment recorded@pepmach marked the first-principles AI finding as false positive, not applicable, or explicitly accepted for
This decision applies only to this commit. A new push requires a new judgment. |
|
/ai-review override fable 3d6c7e2: Clear stopped goal is a text link, not a button; the row holds exactly two buttons. Same ruling as the GPT override on this head. PR owner's layout decision. |
Human judgment recorded@pepmach marked the fable AI finding as false positive, not applicable, or explicitly accepted for
This decision applies only to this commit. A new push requires a new judgment. |
Problem / Motivation
Goal: Replace the goal popover's text buttons with two icon controls, Pause and Play, so a loop's state is readable and one press resumes it.
While a goal loop is running, its popover offers two text buttons, Stop loop and Save. There is no way to pause the loop and pick it up later: the only way to halt it is Stop, which removes the goal for good, and the one-off Trigger nudge sits alone as a text button on the schedule line. A stopped loop and a paused loop look the same, so a reader cannot tell which one is safe to bring back. Bringing a paused or limit-stopped loop back should also give it a fresh budget, so one Play press resumes it instead of being refused until a limit is raised.
Why it matters
A goal loop is the one feature a person leaves running unattended, so its controls have to say what state the loop is in and what one press will do. Today Stop deletes, paused and stopped look alike, and a loop that hit its cycle or time limit cannot be resumed without raising a limit first.
Not a goal
What changed (motivation → approach → change)
Backend (
feat(autonudge): resuming a loop resets its budget). The timer re-checks the cap and budget before every fire, so a loop stopped on either was re-stopped on its first tick afteractive: trueunless the bound was raised._update_unserializedinsrc/kiro_crew/autonudge_service/mutations.pynow runs the user's resume as a fresh run: a revival (keyed onnot was_active) flaggedfresh_runzeroescycle_countand re-anchorscreated_ts. Only the dashboardPATCHroute passes it (dashboard/handlers/autonudge.pyviaautonudge_authz.py); a bare revival keeps count and clock, so amonitor_updateraise buys its increment and the Research Lab and Issue Radar reconcilers (which re-arm any inactive loop every few seconds) cannot turn a cap into a per-run allowance.created_tsis the anchor because every reader of the runtime left measures from it; a running loop's save resets nothing. Also: theautonudge_stateframe (src/kiro_crew/slack/gateway.py) carriesstopped_reasonandnext_due_tsfor every loop, and a reasonless pause landing after a bound keeps the bound.docs/system-specs/modules/learn-cron-dashboard.mddocuments all of it.Frontend (
feat(autonudge): pause and play icon controls in the goal popover).website/src/components/AutoNudgePopover.tsx: while a loop exists the footer is ONE row --Clear stopped goalas a text link at the left on a paused loop (no button chrome) and, at the right, two icon-only controls named inaria-labelandtitle: Pause, and a fire control -- a lightning glyph while running (saves only the edited fields, neveractive, then fires now; a play triangle besideGoal activehad no cold reading), a Play glyph while paused (Pause disabled; Play saves the edits, sendsactive: trueand fires -- one request path, no bound to raise first). A pending edit marks it with a dot and keeps it live while due. The title names the state (Goal active (cycle N/M),Paused, orSet a goalwith no loop) and the line under it is boxed in the state's tone: the countdown in the ok tone while running (neutral while writes are disabled, so it never asserts a fire beside the capability note), otherwisePaused · <why>in the warn tone perstopped_reason(paused by you, stopped by the agent, cycle or time limit reached with the fresh-budget clause, awaiting approval, the last cycle could not run; unknown readsPaused). Clear's confirm replaces the row with a danger-bordered box -- the question in the primary text colour, a filled Clear, a plain Cancel -- under the status line. Gone: the Stopped state label, the Save, Stop and Nudge-now buttons, and every help line naming a tool or a config key.website/src/components/SessionAutomationPopover.tsxkeeps its identity guard andarmedNow;website/src/components/autoNudgeLoop.tsandwebsite/src/hooks/websocket/automationSeed.tschange only comments. Catalogs (website/src/i18n/locales/): sixpaused_*keys added, thirteen retired (the bound help lines, the Save, Stop, Stopped and Nudge-now strings,cycle),en-XAregenerated, dead-key count at baseline.The decision is recorded in
docs/request-for-change/rfc-goal-popover-pause-play-controls.md(state model, controls, resume resets the budget, rejected alternatives; decision owner the PR author, 2026-10-01), indexed there. It ships in this PR, so the First Principles lane, reading status off the base, sees no accepted record yet.Backwards compatibility
Compatible: no route, field or key the wire accepts today is rejected, renamed or removed. The websocket frame gains two fields on plain loops (additive).
update(active=True)on an inactive loop now resets the count and clock -- every caller of that revival (the popover's Play,monitor_update's bound-raise, the Research Lab and Issue Radar re-arms) asks for a fresh run; an active loop's saves are untouched. The retired catalog keys were referenced only by this popover.Tests
test/test_autonudge.py: red-firsttest_revival_runs_a_fresh_budget_and_a_running_save_keeps_the_old_one-- a cap-stopped loop resumes and fires again under the same cap (assert 1 == 2without the fix), a budget-stopped loop resumes withcreated_tsre-anchored, a save on a running loop keeps its count and clock;test_a_bare_revival_keeps_the_count_so_a_reconciler_cannot_mint_a_fresh_allowance(red without the flag) pins the unflagged revival re-stopping unfired;test_a_pause_landing_after_a_bound_keeps_the_boundpins the mirror race.test_monitor_directive_apply.py: a cap raise keeps the count.test_autonudge_handlers_cov80.py/test_autonudge_authz_cov80.py: the route passesfresh_run, the authorizer forwards it and leaves it off by default.test/test_slack_gateway_coverage.py: the frame carriesstopped_reasonandnext_due_tsfor a plain loop.website/src/test/AutoNudgePopover.test.tsx(104 → 71 tests): the paused line per reason as oneit.each(seven reasons plus unknown and absent; never the word Stopped; titlePaused, warn box), the row per state as oneit.each(running, paused, paused at its cap, paused on its time limit: Pause and the fire control, live and named for its press, lightning while running and Play while paused, the dot on an edit), the fire control's write-then-fire on running and paused records (pristine and edited, refused write, refused fire, a legacy stub answer, a due cycle disabling only a pristine press), Pause, Start loop, Clear (text link first in the icon row, the danger-boxed confirm with a filled Clear then Cancel, intentclear, the loops query invalidated, refusal, primed confirm dropped on a record change), the ok box and its neutral tone while writes are disabled. Every behaviour assertion of the previous head is retained.website/src/test/SessionAutomationPopover.test.tsx(98 tests): crew/member mode, the stopped-reason carry per reason, Clear through the bridge, and the two-leg press suite retargeted to Play.AutoNudgePopover.tsxlines 96.25% → 98.08%, branches 91.42% → 91.63%, statements 93.27% → 95.76%;SessionAutomationPopover.tsxlines 96.51% → 96.51%, branches 90.82% → 91.13%;autoNudgeLoop.tsunchanged (96.55% / 95.23%);mutations.py68% → 68% (the two new statements covered, one fewer partial branch),gateway.pyunchanged.Manual verification
Gates on this head (rebased onto main's tip
d8d052f4cc; the full runs below were made on the day's earlier tip5ca51b0a4eand the targeted ones re-run after each rebase): black ratchet, subprocess-encoding, isort, flake8, mypy (2026 files), comment-history, harness parity,scripts/docs-lint.sh; everytest_autonudge*.pyfile plus the gateway coverage and composition-contract files-n0(1,746 tests);scripts/local-gate.pybackend related set 121,901 passed with 48 failures all in the sandbox's own class (test_cloud_source.py× 40'/' is owned by another account (uid 65534), the restart-supervisor and adopted-daemon process tests, two conductor-skill subprocess tests, one served-model test; the twotest_slack_gateway_composition_contract.pyfailures were a relativePYTHONPATHin acwd-changed subprocess and pass with an absolute one);npm run build; the full website suite 2,551 files / 42,866 tests green (the electron stage has noelectronmodule installed on this host);npx eslint src/ --max-warnings 0;vitest run src/i18nandnpm run i18n:checkwithI18N_BASE_REF=origin/main, all ten gates ok. Frames below are the built SPA of this head served with the repository's gateway-free API stub, driven by Playwright, 10/10 states passing their DOM assertions (row names offaria-label, every icon'stitleequal to its name with no visible text, the title and the boxed status line per state in the right tone, the dot only on an edit, Clear as a text link first in the icon row, the danger-boxed confirm with a filled Clear, noStoppedand no tool or config name anywhere); re-shot after the owner's 23:20Z ruling moved the implementation onto the state mocks. No local review lane ran; the server lanes are the first model read.Screenshots / video
Before (
origin/main): text buttons, Trigger nudge on the schedule line, and a paused record indistinguishable from a stopped one.After (this head): the title names the state, the boxed status line under it, Pause and the fire control (lightning while running, Play while paused), Clear as a text link in the same row.
The other states
Related Issues
Supersedes #11478 (the same redesign, reopened on a fresh branch so the review starts from a clean thread). No linked issue: operator product decision delivered through the crew conductor.
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)Contribution License Agreement