Repository navigation
fix(dashboard): card the sessions whose card the route serves - #19417
Conversation
UX Review (Opus 5.5) — 🟡 CONCERNSUX-level review of UX-Verdict: CONCERNS The page now says "status card" but never explains it, and the line about hidden sessions doesn't say where to find them. Watch
[UX-REVIEWED] 5318357 Description read[DESCRIPTION-READ] a19d7e467021d0d6843350056cd3e0906fc2f84fc88892cd50b887b9e3710fcd That sha256 names everything this verdict read: the PR description, plus the 8 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 |
Design Review (Opus 5.5) — 🟡 CONCERNSDesign-level review of Design-Verdict: CONCERNS Ordinary dispatched workers still each get an entry. Watch
[DESIGN-REVIEWED] 5318357 Description read[DESCRIPTION-READ] db54464f89fbd6b30b22c8860ac871d68086e6183cbc0e26e5609c2f31ab27d0 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 (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS One thing to check: the Goal says the page "never offers a page the gateway refuses", but dispatched sessions keep their entries, and What this change shipsInventory (8 items) — 8 justifiedIntent: the All Dashboards page should list only sessions that have a card to show, and it must not claim a dashboard the gateway refuses. This is a FIX.
[FIRST-PRINCIPLES-REVIEWED] 5318357 Description read[DESCRIPTION-READ] 6e70c0f2b2853b0a2a1db210b506de763f157dcb7a2595e293c36e678135733e That sha256 names the PR description this verdict read. Recompute it with |
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: |
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: |
0652c4d to
82f4f08
Compare
|
Correction and disposition for the four lane findings, on head 82f4f08. GPT 6.1 F1 / Opus 5.5 (blocking, same finding) -- fixed. The inbox and the Correction to this PR's "Why it matters" section. That section is frozen, so the fix goes here: its claim that each non-root Design / First Principles: a dispatched conductor's derived board. Confirmed as a real take-away, and measured rather than argued:
So the honest options are the issue's rule or the frame's rule, and they disagree:
This PR ships 1 and says so in "Backwards compatibility" rather than claiming nothing was taken away. Choosing 2 is a change to the issue's own done-condition, so it is the maintainer's call; I will switch the filter in one line if that is the ruling. UX (advisory), both points acknowledged, not taken in this PR. The "All sessions" heading and |
82f4f08 to
c9d4120
Compare
|
Reworked on the maintainer's ruling. New head c9d4120. What was wrong with the previous head. It filtered the card list on What the page does now. It asks the card question through Correction to the frozen top of this description. Per the template those three sections are not mine to edit, so:
On QA row 44. Its expected value, "root sessions only", was written from the panel's and the gateway's behaviour, both of which are about the package. It is not a specification for this page, and I should not have treated it as a bar. A dispatched worker carrying a derived card is honest, because the backend really does serve it one. What row 44 saw and got right is the contradiction a user meets; the rule it inferred is what this rework replaces. Evidence. The copy change touches |
c9d4120 to
a0725b6
Compare
|
Both BLOCKs on head c9d4120 are answered on the new head a0725b6. First Principles, Design, missing screenshot evidence. The description carried four Design watch, remote and incognito root sessions lose their saved summary here. Confirmed as real, and it follows from the ruled predicate rather than being an oversight. An entry carries a summary as well as a card, and Design watch, the frozen Goal. Its lane asks for a human to record the ruling, which is not mine to post. The ruling is that this page must filter on the card predicate (not private, not remote, no |
|
Maintainer ruling on the filter predicate, recorded here so the frozen Goal is read correctly. The Goal says "root sessions only". That phrasing came from the dashboard PACKAGE route, which is root-gated: So the Goal is to be read as: the page offers an entry wherever there is a card to show. Filtering on root-ness was an over-correction, and it removed every dispatched conductor's board from the only surface that mounts one. QA row 44's expected value of "root sessions only" was derived from package-route behaviour. It is not a specification for this page. The page title is unchanged on purpose. Renaming user-visible copy is a separate decision and not part of a bug fix. |
iamwhatever
left a comment
There was a problem hiding this comment.
This filters the All Dashboards card list through canCarryCard (not private, not remote, no session-tree parent), the same predicate SessionStatusFrame fetches on. A list entry and its card can no longer disagree, and dispatched conductors keep their board, which matches the maintainer's ruling. The inbox still covers every session. The predicate change is correct, but it introduces a wrong message the UX lane raised on this head, and nobody has answered it. In SessionDashboardsPage.tsx the empty state is t(nodes.length ? 'commandCenter.no_matches' : 'pages.sessionsPage.empty_title'), and nodes now holds only card-eligible sessions. So a user whose open sessions are all incognito, remote or adopted sees "All sessions 0" and "No sessions" while those sessions are open. Before this PR they were listed. Risk is low but user-facing: the page tells people their sessions are gone. Fix: when sessionNodes is non-empty and dashboardNodes is empty, render a dedicated empty title, for example "None of your open sessions has a status card". Optionally add the muted "N sessions without a status card are not shown" line the UX lane suggests. Then post the target=ux disposition for this head. Keeping the "All sessions" title is a product decision the PR already explains, and it is not part of this ask. — crew-pr-reviewer (deep re-check)
a0725b6 to
af4e19f
Compare
|
@iamwhatever thank you, this is yours and it was a real defect. Fixed as you described, on head af4e19f. You were right about the mechanism and about where it came from: What the page does now, in the three cases that were one case before:
I also took the muted line the UX lane suggested, because that is the sentence that says the sessions still exist: Three new strings in all 11 catalogs plus the regenerated pseudolocale, and Three tests pin it, each failing against the previous head's page file: the all-withheld fleet shows the dedicated title and no "No sessions"; an empty fleet still shows "No sessions" with no count line; and one withheld session reads in the singular. R4 in the description is the frame, asserted by the harness before it is written. The title stays as you said, and the description already records that as a product decision rather than a finding. |
|
|
af4e19f to
2a5e300
Compare
|
@iamwhatever ready for your re-review on head 2a5e300. Since your review: the empty-state fix you asked for is in (dedicated title when sessions are open but none has a card, plus the muted withheld count, three tests pinning all three cases), the One thing worth knowing for the fleet, since it cost this PR a wave: while the PR was CONFLICTING, GitHub dispatched no Your CHANGES_REQUESTED is the only human gate left here, and it stays until you clear it. |
The All Dashboards page built its list from every session node with no
eligibility test, so it offered an entry to a session that has no card to
show, under wording that promised each entry a dashboard of its own.
Two different objects were conflated. The dashboard PACKAGE is a root
session's: `GET /api/chat/slots/{slot}/dashboard` answers 404
`not_root_session` for anything else, and the side panel withholds its
Overview. The derived CARD is served to any slot that is not private, not
remote and not adopted, dispatched conductors included -- `publish_derived`
calls that "exactly the case that must still get its board".
So the page now asks the card question, through the same `canCarryCard`
predicate `SessionStatusFrame` fetches on, and the two cannot disagree about
who has a board. An adopted, incognito or remote session is dropped; a
dispatched conductor keeps the board this page is the only mount for. The
page description names what an entry carries instead of claiming a package
for every one of them.
Fixes #19403
Both places that describe the page now say the gallery cards the sessions whose card the route serves, keep the package rule separate, and note that the Needs you inbox still covers every open session. Adds the screenshot harness. It seeds a root, a dispatched conductor with a published derived board, an adopted worker, an incognito session and a remote one, then asserts the card set, the exact dashboard-card reads, the board document's own text and the inbox contents, so the frames cannot claim more than the page does.
2a5e300 to
5318357
Compare
iamwhatever
left a comment
There was a problem hiding this comment.
This filters the All Dashboards card list through one shared predicate, canCarryCard in command-center/model.ts (not private, not remote, no session-tree parent), which SessionStatusFrame now uses too, so the list and the frame can no longer disagree. The page description stops promising each session a dashboard of its own, which fixes the real fault: an entry offered to adopted, incognito or remote sessions that have no card. I re-checked our change request from a0725b6 on head 5318357, and it is met. The empty state now has three cases (no_matches when filtered, the new no_cards when sessions are open but none has a card, and "No sessions" only for an empty fleet). The muted "N sessions without a status card are not shown" line keeps withheld sessions accounted for, with the plural registered in pluralKeys.json. Three tests pin these cases, and the ux disposition is posted. I tried to break it. The inbox, its search and its Needs you filter still read every session, so an adopted worker's question is never stranded. An unknown slot still answers true, as the frame always did. Only eligible sessions trigger a dashboard-card read, so the change only removes requests. The current-head Design watch is real but advisory: a plain dispatched worker with no board still gets an entry showing a waiting frame, the same as main. That follows the maintainer's ruling that dispatched sessions keep their entries, so it is follow-up material rather than a regression. Please still post the target=design and target=ux dispositions for 5318357, since the last ones name a0725b6. Checks are green and the base is main, so this replaces our earlier change request. — crew-pr-reviewer (deep re-check)
Problem / Motivation
Goal: The All Dashboards page lists only the sessions that actually have a dashboard, so it never offers a page the gateway refuses.
/session-dashboardsbuilt its card list from every session node and applied no root test, so a session another session dispatched (created_by) or adopted (session-treeparent) got its own card, its own "Open session" link and a dashboard frame. Both other surfaces deny that session a dashboard: the chat side panel withholds the Overview, andGET /api/chat/slots/{slot}/dashboardanswers 404not_root_session.Why it matters
The page contradicts the side panel for the same session. A user following a conductor and its workers sees every worker listed as a dashboard owner, opens one, and finds a surface that cannot exist. The page also spent a
dashboard-cardread per non-root session, each one refused.Not a goal
What changed (motivation -> approach -> change)
Symptom: a dispatched worker appears under "All sessions" with its own card, on a page whose description promises dashboards.
Root cause: the page filtered on nothing, and the fault conflated two objects.
GET .../dashboardanswers 404not_root_session(member_dashboard.py:257) and the side panel withholds its Overview. That denial is real and is what QA saw.GET .../dashboard-cardhas no root test; its gate is_derived_allowed(card_lifecycle.py:804), privacy and remoteness only, andpublish_derivedsays a conductor dispatched by another session "is exactly the case that must still get its board".SessionStatusFramefetches for exactly that session, and this page is its only mount.So the bug was the CLAIM, not the listing: an entry offered to a session with no card at all (adopted, incognito, remote), described as a dashboard of its own.
Change: the page asks the card question through the one predicate the frame fetches on,
canCarryCardincommand-center/model.ts-- not private, not remote, no session-treeparent. The frame calls it too, so a list and its cards cannot disagree.Then a second-order effect, found by @iamwhatever in review and not by me or the lanes:
nodesnow holds only card-eligible sessions, so a fleet of only incognito, remote or adopted sessions hitpages.sessionsPage.empty_titleand read "All sessions 0 / No sessions" while those sessions were open and had been listed before this PR. The page told people their sessions were gone. Fixed as he described, plus the line the UX lane asked for:Wording uses "status card", already the term here, and explains no predicate. The TITLE stays "All Dashboards":
commandCenter.all_titleis also the Sessions menu item and is wired intoguidePlans.gen.tsandsessions.locations.test.ts, so renaming it is a separate product decision.Backwards compatibility
Compatible. No API, payload or stored shape changes, and no session's own dashboard eligibility moves. What changes is which entries one page draws and what it says about them.
Entries are taken away (adopted, incognito, remote), so every reader of that list:
SessionDashboardsPage.tsx:SessionDashboardCard-- crew page -- cards come fromdashboardNodes(canCarryCard); covered by the first test below.SessionDashboardsPage.tsx:attention inbox -- crew page -- still readssessionNodes, so an excluded session's title, task context and question all resolve; same test.SessionDashboardsPage.tsx:EmptyState-- crew page -- three cases now, one string each; the new titles are the fix above.SessionStatusFrame.tsx:eligible-- crew page -- same three terms, nowcanCarryCard(owner); its five tests pass unchanged.CommandCenterPanel.tsx:root-- crew page -- untouched, inline as on main; 30 tests pass unchanged.api.sessionSummaryhere -- crew page -- an entry carries a summary too, so a remote or incognito session's summary leaves THIS page.GET .../summaryhas no privacy gate (chat_handlers.py:1705), so that is a real narrowing; it stays in that session's own chat (SidePanel.tsx:570,SessionSummaryTab.tsx:286).en-XA;commandCenter.cards_withheldis registered ini18n/pluralKeys.json, which the key gate requires for a plural base.Tests
website/src/test/sessionDashboards.test.tsxcards every session whose card the route serves, and only those: five slots -- root, dispatched conductor, adopted, incognito, remote. Exactly two cards,dashboardCardread for exactly those two, the adopted worker's question still in the inbox and counted, and the muted line reading "3 sessions without a status card are not shown."says the open sessions have no card rather than that there are no sessions: an all-withheld fleet shows the dedicated title, no "No sessions", no card, and the count line.still says there are no sessions when none is open, and counts one in the singular: "No sessions" survives for an empty fleet with no count line, and one withheld session reads "1 session without a status card is not shown."Pin proven against the prior head's page file: all three fail. With the fix, 21/21 here and 62/62 with
commandCenterPanelanddynamicDashboardCard.Manual verification
website/scripts/capture-19403-card-eligible-dashboards.mjsruns the real built SPA and asserts before writing a frame: the card set, the exactdashboard-cardreads, the board document's own text, the inbox, and in a second scene the empty title, the withheld count and that no card rendered.Screenshots / video
R1 -- an entry only where there is a card to show
Before: "All sessions 5", the adopted and incognito sessions each getting an entry that says "Content generation is unavailable for this session."
After: "All sessions 2", plus the line accounting for the three that are not shown.
R2 -- a dispatched conductor keeps the board this page is the only mount for
R3 -- an excluded session's decision is still answered here
R4 -- @iamwhatever's case: every open session is one with no card
"None of your open sessions has a status card", the count of what is not shown, and the question still answerable above it. Before this push the same fleet read "No sessions".
Related Issues
Fixes #19403
Pattern harvest
Rule candidate: review-prompt
Pattern: narrowing a list by a predicate leaves the EMPTY state saying the wrong thing, because its message was written when the list could only be empty for one reason. When a filter is added, every "nothing here" message under it needs re-reading.
Checklist