Repository navigation
feat(dashboard): the default page shows what a crewmate published - #19657
Conversation
First Principles Review (Opus 5.5) — ✅ PASSPremise-level review of First-Principles-Verdict: PASS Pages that explicitly adopted What this change shipsInventory (9 items) — 9 justifiedIntent: let a crewmate that only calls
[FIRST-PRINCIPLES-REVIEWED] cdc3c91 Description read[DESCRIPTION-READ] 12249257b9b6a5cf0920023d42c3813fe5abd286d9b70b25c313ef0c156b2082 That sha256 names the PR description this verdict read. Recompute it with |
Design Review (Opus 5.5) — ✅ PASSDesign-level review of Design-Verdict: PASS Check that [DESIGN-REVIEWED] cdc3c91 Description read[DESCRIPTION-READ] fb5232e8a54727370cdadb0a2508d70e8122e8e38a70c6be6a103f77e6ce57a4 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) — 🟡 CONCERNSUX-level review of UX-Verdict: CONCERNS The publish-only page shows no publish time, and its only freshness signal is an orange "No new entry" dot that reads as a warning. Watch
Suggestions
[UX-REVIEWED] cdc3c91 Description read[DESCRIPTION-READ] bae1be7aaa660586389fb25272a934606db15704d9620b7a845a019cc6bb9181 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 |
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: |
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: |
38eeb59 to
1fe2724
Compare
1fe2724 to
cdc3c91
Compare
|
rebutted span=639ad9ed80c4
|
|
rebutted span=39464d33f3d4
|
|
rebutted span=ed8a6eeaee6d
|
|
rebutted span=c8bc63363d42
|
|
rebutted (PASS punchline)
|
|
rebutted (PASS punchline)
|
Problem / Motivation
Goal: A crewmate that only calls
panel_publishkeeps showing what it published on its default dashboard page, once the preview flag is gone.Today the published board shows only on the old panel view. The flag-removal PR deletes that view. Without this PR, a crewmate that publishes would show nothing of its board on day one.
Why it matters
Most crewmates adopted no page, so they get the default page,
project-report. Their published boards must not vanish when the tab changes. The data must also stay safe: one slot can hold two crews, and a page must never show another crew's board or a secret.Not a goal
panel_publish, its route, its record or its crew-log entry.What changed (motivation → approach → change)
This implements Decision 2 of the RFC amendment in #19609, and review items 1 and 2 there.
The page reads what the panel route reads. The
panelfold keeps one record per crew, and its top level is just the newest publish, which may be another crew's. The panel read route picks a crew's own record by its ownership digest, and prefers the publish file over the fold, because the file holds every publish while the log can miss one. The page now uses that same rule through one shared helper,owned_panel_recordinsrc/kiro_crew/dashboard/handlers/agent_panel.py, and the route and the page share one ownership check,_owned_by.read_fieldsinsrc/kiro_crew/dashboard/handlers/member_dashboard.pycalls it and hands the record to the feed (page_panelinsrc/kiro_crew/dashboard_feed.py). A crewmate that never published, or a root session's page, gets an empty object. The other crews' records and the digest cannot be bound from a page.The digest stays hidden.
crew_keyjoins_MASKED_VALUE_KEYS, so no page carries it: theagenticandmistakesfolds carry it inside their values too.The stored copy is already clean. The publish route builds the crew-log entry from the scrubbed record, not the raw body. A test now pins that. The page read also redacts again.
The default page draws it.
project-report(src/kiro_crew/dashboard_templates/builtin/project-report/manifest.json, now version 3) bindspublishedto{"fold": "panel", "path": "data"}. A new block insrc/kiro_crew/dashboard_templates/builtin/project-report/template.htmldraws any object as text, like the panel's default template: a short value is a tile, a long one a paragraph, a list of objects a table, a list of values a list, a nested object key/value rows. It wears the same crewmate wrote this tag as every other crewmate-written value on the page. It is hidden when nothing was published. The footer line says "a board published by the crewmate". A page with only a publish says so once, and hides the zero counters above it. A table that drops rows or columns says so. A publish re-reads the open page:panel_publishednow also refetches the dashboard (website/src/hooks/useWebSocket.ts). It usestextContentonly, with no HTML sink.The project-report sample fixtures gain a
panelfold. A newsample_panel.jsonfeeds the page tests, andsample_workstreams-published-only.jsonis the page for a crewmate that only publishes. The builtin README (src/kiro_crew/dashboard_templates/builtin/README.md), the feature map row and the evidence script name the new block.Merge order: this lands after #19609 and before the flag-removal PR (
feat/dashboard-remove-preview-flag), so a crewmate that only publishes keeps its data on day one. If #19335 or #18851 land first, this PR rebases onto them.Backwards compatibility
Compatible: the publish wire contract is unchanged, and a page that never bound
panelreads the same values. Adoptedproject-reportcopies read as stale until re-adopted, as any version bump does.Removes nothing: a page could bind the slot's newest
panelrecord before, and no shipped page did; the only binding is the newpublishedfield.Tests
test/test_dashboard_panel_fold.py(real crew log, fold, file and feed): the page shows the file's publish when the log missed it, like the route; another crew's file on the slug never shows; another crew's publish on a shared slot never reaches this page; a crewmate that never published reads{}and is not stale; a root session reads no panel;ownersandcrew_keydo not resolve from a page; a credential in a published line never reaches the page. Five of six fail on main.test/test_agent_panel_routes.py: the fold holds the scrubbed publish, not the raw body.test/test_dashboard_template_project-report.py: no HTML sink in the page;publishedbindspanel.data.website/src/test/projectReportPage.test.ts: the block draws each shape as text, wears the tag, draws markup as text, hides with no publish, leaves on a refresh that drops it, caps a long table, says when it drops columns, says once that only a publish is shown with no zero counters, and brings the counters back when work arrives.website/src/test/useWebSocket.frameTrace.test.ts: a publish refetches the dashboard (fails without the change). All six fail on the old template.Manual verification
Rendered the page from the fixtures with
scripts/render_dashboard_builtin.py, wide and narrow, dark and light, and with no publish.Screenshots / video
A crewmate with no tracked work that has published a board, the whole page:
A busy page, whole: the published card sits under the tree, and the footer names it.
Light themes
Related Issues
Part of #18642
Checklist
feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)