Repository navigation
fix(chat): say when a chosen steer ran as its own turn - #31055
Conversation
GPT 6.1 Review — ✅ no blocking findingsGPT 6.1 completed its review of This comment is updated in place on each push. Review detailsNo findings. False positive or not applicable? A repository writer can comment: |
|
chore (non-blocking): same-tree rerun for lanes that failed on the CI rate limit c0c49f7 has the same tree as ad20f6f. On ad20f6f, Dependency License Gate, Design Review, First Principles Review, Opus 5.5 Review and UX Review failed between 03:59 and 04:05 UTC with GitHub API rate-limit errors, and PR Readiness failed as their rollup. |
ad20f6f to
c0c49f7
Compare
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Small fix in the right place: the receipt now checks the delivery state the row already has. It adds no new data and takes nothing away. [DESIGN-REVIEWED] 8de3496 Description read[DESCRIPTION-READ] a21e6b93ed1d22ede2910bf4d11de23731b8ad2a84f288e922939946eb55e108 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 |
UX Review (Opus 5.5) — ✅ PASSUX-level review of UX-Verdict: PASS Nothing blocks. The blind reader saw both new receipt sentences and read them correctly: Jev picked interrupt, and the message ran on its own. Notes
[UX-REVIEWED] 8de3496 Description read[DESCRIPTION-READ] 78b037d23aa46bfea2bf6cd01904dbaffd8c45ec985df59812ef96ebd3524931 That sha256 names everything this verdict read: the PR description, plus the 5 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 (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Small fix aimed at the mechanism: the receipt line now checks whether a running turn took the steer. Only the new DOM attribute has no consumer. Notes
What this change shipsInventory (5 items) — 4 justifiedIntent: a sender rating Jev's steer decision sees a receipt that matches what happened to their message. This is a FIX.
[FIRST-PRINCIPLES-REVIEWED] 8de3496 Description read[DESCRIPTION-READ] 164d0adc7b3f0ff764e2c26add62d68aadd1f64cf5ba420075f9c960ff4cffae That sha256 names the PR description this verdict read. Recompute it with |
Opus 5.5 Review — ✅ no blocking findingsReviewed Verdict parsed from the review's SHA-scoped output markers for commit False positive or not applicable? A repository writer can comment: |
c0c49f7 to
3779309
Compare
|
span=03025e01447e Fixed in 3779309. The sentence now names the choice and says that it was not applied, with no second verb. A The stills in the PR body show both rows. A test in |
A steer the running turn did not take runs as its own turn: the turn ended before it read the steer (requeued, then superseded), or the steer path was unavailable and the send fell through to the queue. Its decision receipt still read 'Jev chose: interrupt this turn', which on the row below the Stop card reads as a decision about the new turn. UserMessage now passes ranAsOwnTurn to SteerDecisionLine for a row with no meta.steer or a requeued or superseded steer row. A steer choice on such a row shows 'Jev chose: interrupt the turn, but this runs as its own message' and sets data-ran-as-own-turn. A queue choice is unchanged.
3779309 to
8de3496
Compare
|
span=1ddb32b2e02a Fixed in 8de3496. The receipt now states the choice as an intent, "chose to interrupt", and no longer puts "not applied" next to the verb. On a For the note about frames 1 and 2, each frame caption now says which moment it shows. Frame 2 says that with #18517 the row from frame 1 is hidden when this row lands, so the message shows once. The stills in the PR body are recaptured, and the capture runner fails if a receipt says "not applied". |
KiroCrewCaptain
left a comment
There was a problem hiding this comment.
Tier 1 auto-approve: fix (16 files). Criteria: no conflict, no requested changes, security path denylist clean, design-doc gate clean, security checklist all-NO, no reviewer-lane findings, readiness signal fresh for this head SHA, declared category matches the diff. Category: the steer decision line says when a chosen steer ran as its own message, with new strings in every locale and component tests; display-only change.
A steer the running turn did not take runs as its own turn: the turn ended before it read the steer (requeued, then superseded), or the steer path was unavailable and the send fell through to the queue. Its decision receipt still read 'Jev chose: interrupt this turn', which on the row below the Stop card reads as a decision about the new turn. UserMessage now passes ranAsOwnTurn to SteerDecisionLine for a row with no meta.steer or a requeued or superseded steer row. A steer choice on such a row shows 'Jev chose: interrupt the turn, but this runs as its own message' and sets data-ran-as-own-turn. A queue choice is unchanged. (cherry picked from commit 2bc83d0)
A steer the running turn did not take runs as its own turn: the turn ended before it read the steer (requeued, then superseded), or the steer path was unavailable and the send fell through to the queue. Its decision receipt still read 'Jev chose: interrupt this turn', which on the row below the Stop card reads as a decision about the new turn. UserMessage now passes ranAsOwnTurn to SteerDecisionLine for a row with no meta.steer or a requeued or superseded steer row. A steer choice on such a row shows 'Jev chose: interrupt the turn, but this runs as its own message' and sets data-ran-as-own-turn. A queue choice is unchanged. (cherry picked from commit 2bc83d0)
Problem / Motivation
Goal: The steer receipt on a row that ran as its own turn no longer says "interrupt this turn".
A send in
Auto (Jev)mode while a turn runs can get a steer decision. The user row then shows the receipt "Jev chose: interrupt this turn". The running turn does not always take the steer. If the turn ends before it confirms the steer, the steer row is markedrequeuedand the text runs as its own turn. If the steer path is not available, the send falls through to the queue. In both cases the receipt still says "interrupt this turn". After #18517, the drain copies the receipt onto the new row below the[Stopped]card, and there it reads as a decision about the new turn.Why it matters
The receipt is the only place that tells the sender who decided how the message was handled, and the thumbs beside it rate that decision. On these rows the sentence states an outcome that did not happen, so the sender can rate the wrong thing. This was deferred from the #18517 UX review, span fc2547699fb0.
Not a goal
ambiguous_delivery). That steer is also markedrequeued, and its row carries no flag for the case:STEER_POSSIBLY_DELIVERED_METAis set only on the queue entry, and the drain pops it before it writes the row. The new receipt says only what Jev chose, so it makes no claim about whether the steer was read. The lifecycle line on main still says "Turn ended before this applied" on those rows.What changed (motivation → approach → change)
SteerDecisionLine(website/src/pages/chat/SteerDecisionLine.tsx) picked its sentence fromrecord.choicealone, so a steer choice always read "interrupt this turn". The row meta says whether a running turn took the steer, but the line did not read it.UserMessage(website/src/pages/chat/UserMessage.tsx) now computessteerOwnMessage. It is set when the row has nometa.steer, or whensteerStateisrequeuedorsuperseded. A row with nometa.steerthat carries a steer receipt is the row the drain writes for a requeued steer, or the queued row of a steer that was not available.UserMessagepasses the value to both places that drawSteerDecisionLineas the optionalownMessageprop.For a steer choice on those rows,
SteerDecisionLinestates the choice as an intent, "chose to interrupt", and does not use the "interrupt this turn" command form. There are two sentences, so the row never says the same thing twice:short, "Jev chose to interrupt", on arequeuedsteer row. That row already shows the lifecycle line "Turn ended before this applied, runs as its own message", so the receipt leaves the outcome to that line.full, "Jev chose to interrupt, but this runs as its own message", on every other such row. Those rows have no lifecycle line.The line sets
data-ran-as-own-turn="true"in both cases, anddata-choicestayssteer. The full sentence wraps instead of truncating, and so does the line, so on a narrow pane the score and thumbs move below the sentence instead of cutting off its last clause. A queue choice ignores the prop. A steer row that iswritten,consumedor has no state keeps "interrupt this turn".Two earlier versions failed the UX review. "Jev chose: interrupt the turn, but this runs as its own message" read as a contradiction, because "interrupt the turn" sounds done and "but" then undoes it, and it repeated the lifecycle line on a requeued row. "Jev chose: interrupt, not applied" also read as "it interrupted and it did not", because the verb stood next to its own cancellation. "Chose to interrupt" names the intent, so the outcome after it does not undo a verb.
The two keys,
steer_chose_steer_requeuedandsteer_chose_steer_own_message, are added next tosteer_chose_steerinen.manual.jsonand in the 11 translated catalogs, anden-XA.jsonis regenerated withnpm run i18n:pseudo.The change works on main alone:
requeuedsteer rows and fall-through rows already carry the receipt there. #18517 adds thesupersededstate and copies the receipt onto the drained row. This change readssupersededalready, so either PR can merge first and neither needs a change for the other.Backwards compatibility
Compatible: the prop is optional and unset by default, and rows on disk keep their meta. Only the sentence on rows that run as their own message changes.
Removes nothing: the change adds two sentences and one data attribute and keeps every existing sentence, key and attribute.
Tests
website/src/test/SteerDecisionLine.test.tsx, new describe block "a chosen steer the running turn did not take", eight tests:requeuedsteer row shows "Jev chose to interrupt", setsdata-ran-as-own-turn="true", keepsdata-choice="steer", says "its own message" once for the whole row, and does not say "not applied".supersededsteer row, which draws no lifecycle line, shows the full sentence.written,consumedor stateless steer row keeps "Jev chose: interrupt this turn" with no new attribute.requeuedand aconsumedrow.SteerDecisionLinewithownMessage="full"and a queue record keeps the queue sentence.The requeued, superseded, plain-row and wrap tests failed on the previous head. The others pin behaviour that does not change, so they pass on both sides.
Proof: not provable -- frontend test (prove.py runs Python only)
Manual verification
N/A. Unit coverage is sufficient, and the stills below draw the real components.
Screenshots / video
The sheet draws three separate moments with the real
UserMessageandStopEventCard, and each frame has its own caption. Frame 1 is the steer row markedrequeuedabove the Stop card, before its own turn starts: its lifecycle line says the text runs as its own message, and the receipt only says what Jev chose. Frame 2 is the same message later, on the row its own turn runs from, below the Stop card. With #18517 the row from frame 1 is hidden at that point, so the message shows once. That row has no lifecycle line and gets the full sentence. Frame 3 is a consumed steer for comparison, which keeps "interrupt this turn". The capture runner asserts the sentence and thedata-ran-as-own-turnattribute in each frame, that frame 1's receipt does not repeat "own message", and that no receipt says "not applied".Related Issues
Deferred from the #18517 UX review. Tracked first as Pearcekieser#46.
Sibling sweep
Swept: git grep -n "SteerDecisionLine|steer_chose_steer" -- website/src
Sibling: website/src/pages/chat/UserMessage.tsx -- changed here
Sibling: website/src/pages/chat/DecisionStrip.tsx -- not affected -- the skills strip names no delivery outcome
Pattern harvest
Not generalizable: the receipt line is the only surface that states a decision next to a delivery outcome it does not read, and the change makes it read that outcome.
Checklist
Contribution License Agreement
Contributions are accepted under the project's Apache-2.0 license.
Fixes #31054