Skip to content

feat: root session side panel shows its own Dynamic Dashboard - #18105

Merged
iamwhatever merged 2 commits into
mainfrom
feat/18022-root-session-dashboard
Oct 12, 2026
Merged

iamwhatever merged 2 commits into
mainfrom
feat/18022-root-session-dashboard

Conversation

@chenmingwei23

@chenmingwei23 chenmingwei23 commented Oct 8, 2026 •

Copy link
Copy Markdown
Contributor

Problem / Motivation

Goal: A root session's chat side panel shows that session's own template-based Dynamic Dashboard, and the two older dashboard surfaces go away.

Three surfaces do one job today. A crewmate's Dashboard tab is composed from declared fields plus a shipped template (CrewDynamicDashboard, GET /api/members/{slug}/dashboard). The chat side panel's Dashboard is HTML the agent writes itself (TaskDashboardFrame, a task-dashboard artifact). The status-tile HUD above the composer (CommandCenterDock, StatusTiles, TileList) is a third, hand-built readout. A session gets no template-based page at all.

Why it matters

A model-written page spends tokens on every refresh and can state a number the crew log does not support; the declared-field path cannot. The surfaces drift, so the same work reads differently depending on where you look. The tiles take vertical space above the composer in every session, including worker and subagent sessions that have nothing session-level to report.

Not a goal

  • No new template, and no change to any template's fields or folds. project-report only gains session wording for its heading.
  • No dashboard for a dispatched, adopted or subagent session.
  • No change to the crewmate Dashboard tab's behaviour or route.
  • No change to the automatic session card (SessionStatusFrame) or the all-session page beyond dropping its published-view selector.
  • The preview flag is not turned on by default.

What changed (motivation → approach → change)

Approach: reuse the declared-field dashboard that already serves a crewmate, keyed by a root session's slot instead of a member slug, rather than building a fourth surface.

  • Read route. GET /api/chat/slots/{slot}/dashboard (member_dashboard.api_session_dashboard) shares the member route's body (_serve, _render, _safe_body, _page_safe redaction). It is owner-only and denies app callers, like the member route. Root is the existing card_lifecycle.is_root_session(slot), with no new root test; a dispatched or adopted session is answered 404 not_root_session. Every fold a field reads comes from that slot (read_fields(..., slot=)). The route is bound through the deferred binder.
  • Store. instance.session_instance_key(slot) files a session's instance under session-dashboards/<sha256[:32]>, beside the member directories. A key not exactly that shape is refused before it reaches a path.
  • Shared templates. One built-in catalog serves crewmates and sessions; no per-surface copies. The page payload carries subject (crewmate | session), so project-report reads "What this session is doing" on a session page (English only; the zh-CN string waits on feat(dashboard): render built-in dashboard pages in the reader's UI language #18014).
  • Dashboard tool caller gate (security-relevant). _resolve_publishing_crew in src/kiro_crew/dashboard/handlers/agent_panel.py gains a root_session branch, passed only by _resolve_dashboard_caller (the dashboard_fields / write / templates / preview / apply / rollback routes). It runs after every existing gate: internal secret, agent.crew_panel, app denial, strict session identity, restricted mode, slot check. A caller bound to no crew then resolves to its OWN slot's store key, and only when is_root_session(slot) holds. Otherwise it gets 403 not_root_session, audited like every other refusal. A slotless subagent still gets no_dashboard_slot. The crew webview publish route does not take the branch and still answers no_crew. An unseeded session tree fails closed. Session writes broadcast {slot}.
  • Replies reach the composer. CommandCenterPanel passes onAct to the session dashboard; SidePanel forwards it from ChatPage, which merges the reply into this chat's composer and focuses it. Nothing is sent until the person sends it.
  • No-dashboard line. A dispatched or adopted session shows one muted line, commandCenter.no_dashboard_here, in place of the missing Overview.
  • Side panel. CommandCenterPanel's Overview mounts CrewDynamicDashboard with a session target. The component takes DashboardTarget = member | session rather than being copied. A non-root session loses the Overview segment and opens on Questions.
  • Removed: Bolin's status-tile HUD above the composer (CommandCenterDock.tsx, StatusTiles.tsx, TileList.tsx and their wiring in ChatPane.tsx / ChatPage.tsx), and the agent-published task-dashboard Overview (TaskDashboardFrame.tsx, commandCenter.prompt.ts / REQUEST_PUBLISHED_VIEW, the task-dashboard artifact source in useCommandCenter, the all-session page's published-view selector). Their tests, 26 dead commandCenter.* i18n keys and the dock capture script went with them.
  • Flag. Everything stays behind the existing PREVIEW_DASHBOARD feature preview.
  • Agent prompts unchanged. The Crew Mode prompt in agent.py and the artifacts, goal-conductor and pipeline-conductor skills still describe publishing a task-dashboard artifact. Those prompts reach every user whether or not the preview is on, so changing them is left out of this PR. Retiring that text is a follow-up for after PREVIEW_DASHBOARD defaults on.

RFC: docs/request-for-change/rfc-crewmate-dynamic-dashboard.md (in-progress) records the declared-field dashboard. It does not record removing the HUD and the published Overview. That decision was taken on #18022 by the issue's owner.

Backwards compatibility

Compatible: every route that answers today still answers the same body. The new route and the store directory are additive. The gate change only admits a caller that was refused no_crew before, and only to its own slot.

Reader: website/src/components/ChatPane.tsx:CommandCenterDock -- chat -- render and onOpenCommandCenter prop removed; ChatPage.dockSearchClose.test.tsx asserts no dock renders
Reader: website/src/pages/ChatPage.tsx:CommandCenterDock / openCommandCenter -- chat -- removed; the side-panel tab is still opened from the + menu
Reader: website/src/pages/chat/command-center/CommandCenterPanel.tsx:TaskDashboardFrame, REQUEST_PUBLISHED_VIEW -- chat -- replaced by CrewDynamicDashboard; commandCenterPanel.test.tsx
Reader: website/src/pages/chat/command-center/SessionDashboardsPage.tsx:TaskDashboardFrame -- crew page (/session-dashboards) -- selector removed, automatic card kept; sessionDashboards.test.tsx
Reader: website/src/pages/chat/command-center/SessionStatusFrame.tsx:TASK_DASHBOARD_SANDBOX -- chat -- constant moved into this file
Reader: website/src/pages/chat/command-center/useCommandCenter.ts:TASK_DASHBOARD_TAG, dock, relevant, finished -- chat -- only the dock read them; commandCenterSources.test.tsx
Reader: website/src/i18n/locales/*.json:commandCenter.* (26 keys) -- release -- no remaining source reference; commandCenterRetired.test.ts guards the names
Reader: website/src/utils/storageGc.ts:mc-task-dashboard-dismissed: -- chat -- kept so old browsers still collect it

Tests

  • test/test_session_dashboard.py (new, 19):
    • root gets its page; dispatched and adopted get 404 not_root_session; an unknown slot gets 404;
    • the slot read is owner-only, app-denied and redacted;
    • the route is bound through the deferred binder; a store key cannot name a member path;
    • a root session writes its own page (the preview link is the slot route; apply lands on its own key; the read shows it);
    • dispatched and adopted callers are refused the write and nothing is staged;
    • a slotless subagent is refused; a root cannot publish a crew webview; the internal secret is still required; an unseeded tree refuses;
    • a session page is told subject: session, a crewmate page crewmate, and project-report switches its heading.
  • Updated: test_agent_panel_routes (the resolver refusal count goes from 7 to 8), test_member_dashboard_routes (structural test admits the session read), test_dashboard_frame (payload shape gains subject).
  • vitest:
    • commandCenterPanel covers root vs worker vs adopted, no mount while inactive, and questions kept out of the Overview;
    • CrewDynamicDashboard covers the session target reading the slot route and invalidation by slot;
    • new commandCenterRetired checks that no source names the removed HUD or published-Overview parts;
    • commandCenterSources, sessionDashboards, ChatPage.dockSearchClose, statusStackLayering, Glass.thickness, MembersPage, CrewDashboardTab and storageGc were trimmed to match.
  • Locally on the final tree: the related pytest (668 agent-prompt, skill, dashboard and route-table tests; 835 earlier across the dashboard modules) and vitest (15 files) passed; tsc, eslint, flake8 and semgrep were clean.
  • One mutation per change, each caught by a test: 14/14 (slot frame sent to owners only, root check, owner gate, app denial, redaction, slot folds, gate accepts any session, publish takes the root branch, session preview link, UI non-root Overview, session target route, slot invalidation, HUD name reintroduced, session subject).

Manual verification

Pod for this branch with the preview on. Seeding:

  • a root slot whose crew log holds a goal and four work items;
  • an adopted slot, with a session-tree parent edge to the root;
  • a dispatched slot, with created_by set to the root.

What the shots show:

  • The root's side panel draws project-report from its own ledger: "Ship the 0.9 release", 3 in flight, 1 of 4 accepted, the blocked item under Needs you.
  • The dispatched and adopted sessions show only Questions and Approvals, and their API answers 404.
  • No tiles render above the composer, and the DOM has no command-center-dock or status-tiles.

Screenshots / video

Root session: the Overview is its own Dynamic Dashboard, the heading reads "What this session is doing", and nothing sits above the composer.

Root session side panel showing project-report from its own work ledger

Root session: clicking a reply option on the page ("Re-dispatch it") puts that reply in this session's composer, where the person sends it.

A dashboard reply option filled into the session composer

Dispatched worker session, on this head: no Overview segment. The muted line reads "This session's work shows on the dashboard of the session that started it." The session's created_by names the root.

Dispatched session panel showing the no-dashboard line

One /session-dashboards card after the change: the published-view selector and its "no published view" line are gone. The card keeps its status, the automatic card notice and the saved summary.

One session-dashboards card after the change

The earlier stills of the dispatched and adopted sessions came from a revision before the no-dashboard line, and they show only Questions and Approvals. The two stills above replace them as evidence for this head.

Roster read forced to fail (/api/members 500): the Overview shows one ErrorNotice above where the page would be, and mounts no page.

Overview when the crewmate roster read fails: one error notice, no page

Overview chunk forced to fail (CrewDynamicDashboard chunk aborted): the boundary shows the same notice, and the Questions and Approvals tabs stay usable.

Overview when its chunk fails to load: error notice, Questions tab still there

Related Issues

Fixes #18022

Follow-up, after PREVIEW_DASHBOARD defaults on: retire the task-dashboard prompt text in agent.py and the three skills.

Sequenced with #18014 (dashboard i18n); the zh-CN heading string lands after it.

Follow-ups

Out of scope here, tracked for later:

  • Retire the task-dashboard prompt text in agent.py and the artifacts, goal-conductor and pipeline-conductor skills, after PREVIEW_DASHBOARD defaults on.
  • With KIROCREW_CREW_LOG off the session tree never seeds, so is_root_session answers false and every root session's Overview gets 404 not_root_session while the panel still offers it. Align the client and server root tests.
  • UX: on a session page the project-report title stays "Project report". Using the session title needs a new payload field, redaction and template code, so it is more than a one-line change.
  • UX: give dashboard_none_yet session wording.
  • UX: "Needs you" appears four ways on a root session's panel (the Questions header, and the template's count card, section and summary). Fixing it means dropping the template's count card on session pages or linking it to the Questions tab, which is more than a heading rename.
  • UX: the page's reply options only fill the composer; add a line near them saying nothing is sent until the person sends it.
  • UX: the no-dashboard line on a dispatched or adopted session should name the session that started it and link to it through onOpenSession.
  • First Principles: replace pinned_fs.create_and_open_dir_at and the two _refuse_linked_* helpers in instance.py with pinned_fs.open_pinned_descendant_dir(data_home(), ("session-dashboards", digest), create=True), keeping the race and junction tests.
  • E2E session-tags-e2e.spec.ts 17b fails in CI on unrelated PRs on the same main base too (e.g. fix/crew-warm-greeting, feat/default-template-picker-18411, feat/mate-first-crewmate, fix/publish-visibility-p0). Locally the spec passes alone and as a whole file on both origin/main 3387bdd and this branch. In the CI trace, rowA.click scrolls row A to the top of the session list, so row B sits under the sticky search bar and the backdrop right-click at y=105 lands on the bar instead of row B. Left for a separate fix of the spec geometry.

Checklist

  • At most two commits (one is the norm), with a Conventional Commits title (feat|fix|docs|style|refactor|perf|test|chore|ci|build|revert: ...)
  • Existing tests pass and new tests added for new functionality
  • Self-review completed; code follows project style guidelines
  • Documentation updated (if applicable)
  • No secrets, credentials, or internal references in the diff

Contribution License Agreement

Contributions are accepted under the project's Apache-2.0 license; see the Licensing section of CONTRIBUTING.md.

@chenmingwei23
chenmingwei23 requested review from a team and KiroCrewCaptain as code owners October 8, 2026 09:06
@github-actions github-actions Bot added the readiness: checking Automated validation is still running label Oct 8, 2026
@github-actions

github-actions Bot commented Oct 8, 2026 •

Copy link
Copy Markdown
Contributor

GPT 6.1 Review — ✅ no blocking findings

GPT 6.1 completed its review of a193a00991d23d1f3af918b3f7b7085fabb383c4 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] a193a00

False positive or not applicable? A repository writer can comment:
/ai-review override gpt a193a00991d23d1f3af918b3f7b7085fabb383c4: <one-sentence reason>

@github-actions

github-actions Bot commented Oct 8, 2026 •

Copy link
Copy Markdown
Contributor

Design Review (Opus 5.5) — 🟡 CONCERNS

Design-level review of a193a00991d23d1f3af918b3f7b7085fabb383c4 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: CONCERNS

Session pages are filed by slot key, and slot keys are reused, so a new conversation in a reused slot inherits the previous conversation's dashboard.

Watch

  • Store identity is the slot, not the session. The patch keys the store as session_instance_key(slot.key) → session-dashboards/<sha256[:32]>, and docs/system-specs/modules/session.md says plainly that "slot keys ARE reused". No hunk deletes or resets that directory when a slot is closed, reset or recreated. So the next root conversation in that slot opens on the old chosen template and the old agent-written values and written_at. Under the design's own rule, every number on a session's page comes from its own crew log, so showing another conversation's values breaks that rule. The directories also pile up with no reaper. The layout is persisted on disk, so changing the key after the preview ships means a migration.
    Clears when: the key includes the session identity (for example slot plus history/session id), or close_slot/reset clears the slot's instance; plus a test that a recreated same-key slot reads no prior page.
  • The token cost the PR is meant to remove is still there. "Why it matters" says a model-written page "spends tokens on every refresh". But the Crew Mode prompt and the conductor skills still tell agents to publish task-dashboard, and with this PR nothing in the side panel shows that artifact any more. Preview users now pay that cost for a page they can't see. The description defers this on purpose.
    Clears when: a tracked follow-up issue removing the task-dashboard publishing text is linked from the PR, tied to PREVIEW_DASHBOARD defaulting on.

[DESIGN-REVIEWED] a193a00

Description read

[DESCRIPTION-READ] df287559b6eea3392f94a180ef8f91684f81a3ec90e4281eea0a064c99f0a40d

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 .github/scripts/pr-description-capture.sh over the same inputs; if it differs, the description or that evidence changed after this verdict was formed and any finding drawn from either is unproven.

@github-actions

github-actions Bot commented Oct 8, 2026 •

Copy link
Copy Markdown
Contributor

First Principles Review (Opus 5.5) — 🟡 CONCERNS

Premise-level review of a193a00991d23d1f3af918b3f7b7085fabb383c4 — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

First-Principles-Verdict: CONCERNS

The session store adds a second, pinned-descriptor write path. Member pages, written by the same gateway into the same data home, still use the plain path.

Not justified as shipped

  • Item 7 — oversized: the threat in the _refuse_linked_session_dir docstring ("A root session's agent runs in a sandbox that can create links inside the data home… would have the gateway create, lock and write files wherever it points") is not specific to sessions. Yet _locked/_commit/_write only pin when directory.parent == data_home() / _SESSION_DASHBOARDS_DIR. That makes 1 of 2 store kinds pinned. _PINNED_DIR plus fsync_open_dir and create_and_open_dir_at exist only for that one branch, so instance.py now carries two write paths for one store.
    Subtraction: drop the session-only pinned branch (_PINNED_DIR, _pinned_session_dir, the pinned arms of _commit/_prune_versions, fsync_open_dir, create_and_open_dir_at) so session pages use the path member pages already use.
    Clears when: the author shows a link-planting reach that applies to session-dashboards/ but not to member directories, or the pinned branch is removed.

What this change ships

Inventory (8 items) — 7 justified

Intent: give a root session its own template-based dashboard in the chat side panel and retire the two overlapping preview surfaces. This is an ADDITION.

  1. A root session's side-panel Overview shows its own template dashboard (CrewDynamicDashboard, session target) — justified
  2. New read route GET /api/chat/slots/{slot}/dashboard; owner-only; returns 404 for non-root sessions — justified
  3. Dashboard MCP tools let a root session with no crew write its own page; others get 403 not_root_session — justified
  4. Dispatched or adopted sessions lose the Overview, open on Questions, and show one muted line — justified
  5. The status-tile dock above the composer is removed (it was only shown with PREVIEW_DASHBOARD on, which defaults off) — justified
  6. The agent-published task-dashboard Overview and the All Dashboards page's published-view selector are removed (same preview gate) — justified
  7. Session pages are stored under session-dashboards/<digest> through a separate pinned-fd write path — oversized
  8. Pages get a subject value, so project-report headings say "this session" — justified

[FIRST-PRINCIPLES-REVIEWED] a193a00

Description read

[DESCRIPTION-READ] f21bd8b8d73a46d84361fab3b07e1dd8fa35d7d8bda28b641a73109804e18126

That sha256 names the PR description this verdict read. Recompute it with .github/scripts/pr-description-capture.sh; if it differs, the description changed after this verdict was formed and any finding drawn from the description is unproven.

@github-actions

github-actions Bot commented Oct 8, 2026 •

Copy link
Copy Markdown
Contributor

UX Review (Opus 5.5) — ✅ human override accepted

@chenmingwei23 overrode this lane for a193a00991d23d1f3af918b3f7b7085fabb383c4 via /ai-review override (this commit only).

@github-actions

github-actions Bot commented Oct 8, 2026 •

Copy link
Copy Markdown
Contributor

Opus 5.5 Review — ✅ no blocking findings

Reviewed a193a00991d23d1f3af918b3f7b7085fabb383c4 — this comment is updated in place on each push.

Review details

No findings that block the merge; one advisory finding below.

FINDING — src/kiro_crew/dashboard/handlers/agent_panel.py:297 — if not crew_name and root_session: is never reached from an ordinary chat tab. An empty slot.agent resolves to config.default_agent with selection_kind="member" (_resolve_agent_selection → crew_alias = bindings.resolved_alias in chat_runner → capability_member), so get_agent_selection returns ("member", <default agent>). As a result, dashboard_write/apply/preview land on members/<default>/dashboard and broadcast {slug}. Meanwhile CommandCenterPanel finds no roster row with slot_key === slot and mounts {kind:'session'}, so the new Overview never shows anything the agent writes. The tests miss this because they stub the selection as ("template", "kiro"). → Fix: have the root-session branch decide from the caller's slot, not from the member selection: a member selection whose slot is not that member's own DM slot (_dashboard_slot) should resolve to the session key, matching the client's roster slot_key test.

[OPUS-REVIEWED] a193a00

Verdict parsed from the review's SHA-scoped output markers for commit a193a00991d23d1f3af918b3f7b7085fabb383c4.

False positive or not applicable? A repository writer can comment:
/ai-review override fable a193a00991d23d1f3af918b3f7b7085fabb383c4: <one-sentence reason>

@github-actions github-actions Bot added readiness: action required A blocking check or review needs attention and removed readiness: checking Automated validation is still running labels Oct 8, 2026
@chenmingwei23
chenmingwei23 force-pushed the feat/18022-root-session-dashboard branch from 7cf5a65 to 6866edb Compare October 8, 2026 09:57
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: action required A blocking check or review needs attention and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Oct 8, 2026
@chenmingwei23
chenmingwei23 force-pushed the feat/18022-root-session-dashboard branch from 6866edb to dac0c26 Compare October 8, 2026 11:06
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision and removed readiness: action required A blocking check or review needs attention readiness: checking Automated validation is still running labels Oct 8, 2026
@iamwhatever

Copy link
Copy Markdown
Contributor

This PR gives a root session's chat side panel its own template-based Dynamic Dashboard (new GET /api/chat/slots/{slot}/dashboard, a session-dashboards/ store). It also deletes the status-tile HUD and the agent-published task-dashboard Overview. I read the description, the production diff summary and the current-head lanes (dac0c26). CI is green, and GPT reports no findings. I am holding instead of approving for three reasons. First, it changes a security-relevant caller gate: _resolve_publishing_crew in src/kiro_crew/dashboard/handlers/agent_panel.py gains a root_session branch that admits callers who were refused no_crew before. Second, it is a large product/UX change that removes two surfaces, with +990/-2270 across 64 files. Third, three whole-design lanes are CONCERNS and have no disposition yet. Open points worth settling first: Design and First Principles both note that agent.py:4480 and the artifacts, goal-conductor and pipeline-conductor skills still tell agents to publish a task-dashboard artifact that nothing renders now, which leaves the token-cost harm in place. Opus notes that in a crewmate-bound root slot, api_session_dashboard (member_dashboard.py:263) serves the session instance while the agent's writes go to the member slug, so the Overview never shows what the agent installed. Opus also notes that slot_projection in useWebSocket.ts:657 does not invalidate session-dashboard, so fold numbers lag by up to 60 s. A maintainer should own the gate change and the surface removal. — crew-pr-reviewer

@github-actions github-actions Bot added the merge conflict Branch has merge conflicts with its base — author must resolve before merge label Oct 8, 2026
@chenmingwei23
chenmingwei23 force-pushed the feat/18022-root-session-dashboard branch from dac0c26 to c0d0a44 Compare October 8, 2026 20:52
@github-actions github-actions Bot added readiness: checking Automated validation is still running readiness: passed Eligible automated validation passed for the current revision merge conflict Branch has merge conflicts with its base — author must resolve before merge and removed merge conflict Branch has merge conflicts with its base — author must resolve before merge readiness: passed Eligible automated validation passed for the current revision readiness: checking Automated validation is still running labels Oct 8, 2026
@chenmingwei23

Copy link
Copy Markdown
Contributor Author

Both blocking findings fixed in d941db4.

CommandCenterPanel.tsx:149 — Overview chunk failure discarding an unsent answer. The Overview is a lazy import; a failed chunk load (e.g. a 404 after a frontend rebuild with the tab open) rejected inside the Suspense, climbed to SidePanel's outer boundary, and unmounted the whole panel — taking the answer draft the Questions tab holds in local state. The lazy Suspense now has its own dashboard-only ErrorBoundary (retryOnly, fallback is an ErrorNotice), so the failure stays contained in the Overview region and the cards and their drafts stay mounted. New test: "contains an Overview chunk-load failure instead of unmounting the panel".

CommandCenterPanel.tsx:80 — a failed roster refresh hidden by a cached DM row. rosterFailed was !dm && roster.isError, so a successful DM read followed by a failed refresh left dm non-null, suppressed the notice, and showed the cached page with no sign the request failed. It is now rosterFailed = roster.isError: the notice renders on the failure while target keeps the cached DM dashboard mounted. New test: "says a refresh failed while keeping the cached DM dashboard mounted".

Local: tsc, eslint, i18n lint, and the full commandCenterPanel.test.tsx suite pass on the pushed tree.

@chenmingwei23

Copy link
Copy Markdown
Contributor Author

/ai-review override ux a457e1a: maintainer asked not to commit binaries; screenshots are linked in the description from branch screenshots/18022; collector cannot read them

@chenmingwei23

Copy link
Copy Markdown
Contributor Author

Rebased onto current main and pushed as a457e1a283. The only conflict was the import block in website/src/components/MarkdownRenderer.tsx; I kept both sides' symbols that the file body uses (useSessionChip, ClosedSessionNotice, dashboardPreviewTargetFromHref) and dropped the two that nothing references. Locale catalogs merged by key union.

On the rebased tree: isort/black/flake8/mypy clean on the changed Python; the backend tests for this change pass (including the dashboard server composition contract, so the MCP route table is unchanged); tsc and eslint clean; the 16 touched vitest files pass (615 tests); the static i18n checks and gen-ui-index --check are green.

UX Review disposition: the BLOCK is "cannot evaluate: missing the Overview error states" — the two new states (command-center-roster-error and command-center-overview-error) are transient failure UIs added in this change and are not in the four linked stills, which the reviewer reads from the description. Per the no-committed-binaries rule the images stay off the PR branch, so I have posted a UX override on this head for the evidence gap only.

The two "Watch" items (the no-dashboard line on a non-root session should name and link the starting session; the reply inserted into the composer should drop the internal id) and the "Suggestions" item are advisory on this verdict and are tracked in the Follow-ups section of the description, not folded into this push.

@github-actions

Copy link
Copy Markdown
Contributor

Human judgment recorded

@chenmingwei23 marked the ux AI finding as false positive, not applicable, or explicitly accepted for a457e1a2836ad3b85fb43eb7936ec2f6ae67392a.

maintainer asked not to commit binaries; screenshots are linked in the description from branch screenshots/18022; collector cannot read them

This decision applies only to this commit. A new push requires a new judgment.

buluoray
buluoray previously approved these changes Oct 11, 2026

@buluoray buluoray left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Approving on a457e1a. This supersedes my earlier comment review on 994229c.

What was holding it, and where it stands on this head:

  • Design sign-off -- posted on #18022.
  • Spec drift -- fixed. pipeline-conductor.md now says no panel reads a task-dashboard artifact and that the Overview is the session's catalog page, which matches artifacts.md.
  • Staged session preview (iamwhatever's second point) -- fixed. dashboardPreview.ts adds SESSION_PREVIEW_PATH_RE for /api/chat/slots/<slot>/dashboard, CrewDynamicDashboard passes staged into api.sessionDashboard, and dashboardPreview.link.test.tsx pins the session link, the round-trip and the side-panel open.
  • Crewmate DM slot (third point) -- fixed. CommandCenterPanel.tsx:87-89 mounts the member target when the roster binds this slot to a crewmate, mounts nothing while the roster is pending, and shows an error instead of guessing when the roster read fails. commandCenterPanel.test.tsx covers the DM page and both failure cases.

Checked against source on this head; I did not run the suites locally. At review time 88 checks passed, 8 skipped, 10 pending (review lanes and readiness), 0 failed.

Non-blocking, still open from my earlier pass: _writer_key has no test, and is_root_session returning True when the projection raises would read safer as fail-closed on the write gate.

@chenmingwei23

Copy link
Copy Markdown
Contributor Author

/ai-review override ux a9109df

@github-actions

Copy link
Copy Markdown
Contributor

AI-review override not recorded. Use /ai-review override <fable|gpt|design|ux|first-principles|scope|all> <current-sha>: <one-sentence reason>.

@chenmingwei23

Copy link
Copy Markdown
Contributor Author

/ai-review override ux a9109df: Rebase-only head with no UI change since the human-accepted a457e1a; the error-state screenshots are embedded in the PR body.

@github-actions

Copy link
Copy Markdown
Contributor

Human judgment recorded

@chenmingwei23 marked the ux AI finding as false positive, not applicable, or explicitly accepted for a9109dff967de6d20b4f9f2b2e840b8f20f29d97.

Rebase-only head with no UI change since the human-accepted a457e1a; the error-state screenshots are embedded in the PR body.

This decision applies only to this commit. A new push requires a new judgment.

@chenmingwei23

Copy link
Copy Markdown
Contributor Author

Re: review 2026-10-10. Per-point status on the current head a9109dff967de6d20b4f9f2b2e840b8f20f29d97. Every point maps to a commit or a concrete reason; nothing new to push.

1. pipeline-conductor spec (buluoray's open point). Fixed in a868acb844. docs/system-specs/modules/pipeline-conductor.md no longer says the task-dashboard artifact shows "in the shared chat/Crew Dynamic Dashboard". The Components section now reads "No panel reads a task-dashboard HTML artifact" and explains that a root session's side-panel Overview is that session's own catalog page, read from the pipeline's own work ledger and crew log — not a new status store. The skill text that still names the artifact is retired in a separate follow-up.

2. Session preview link. Fixed in a868acb844. website/src/utils/dashboardPreview.ts now carries SESSION_PREVIEW_PATH_RE for /api/chat/slots/<slot>/dashboard alongside the existing /api/members/<slug>/dashboard match, and resolves a session href to a dashboard-preview:session:<slot> ref that no member slug can spell. api.sessionDashboard takes a preview flag and sends preview=1. CrewDynamicDashboard passes the staged flag through for a session target (api.sessionDashboard(target.slot, locale, staged)), so a person checking a staged session page before apply gets the rendered preview, not raw JSON. Covered by dashboardPreview.link.test.tsx, CrewDynamicDashboard.test.tsx, and DashboardPreviewPanel.test.tsx.

3. Crewmate DM slot gets the member target. Fixed in a868acb844. CommandCenterPanel looks the slot up in the roster by slot_key; a crewmate DM row resolves to a member target ({ kind: 'member', slug, member }) so it shows the page its agent writes, while a plain root session resolves to a session target. Nothing mounts until the roster answers, and a roster read failure is surfaced rather than guessed around (so a DM is never shown an empty session page). The later head a9109dff96 only hardens the surrounding error handling (Overview chunk-load failure containment and roster-refresh error display); it does not change this target selection.

4. Dispositions for Design, First Principles and UX. All three checks are green on this head: Design Review, First Principles Review and UX Review all report success, as do GPT 6.1 Review, PR Scope, Security scope candidates (Opus 5.5), and Publish readiness. The First Principles note that create_and_open_dir_at duplicates open_pinned_descendant_dir is a simpler-shape observation, not a defect; it is left as a follow-up to keep the reviewed link-refusal code and its tests stable. The UX watch items (naming the starting session on the no-dashboard line, the session-title heading, and the "nothing is sent until you send it" note on reply options) are tracked as follow-ups.

Head stays a9109dff967de6d20b4f9f2b2e840b8f20f29d97; all checks green and Publish readiness passing.

iamwhatever
iamwhatever previously approved these changes Oct 11, 2026
@chenmingwei23

Copy link
Copy Markdown
Contributor Author

/ai-review override ux 7fef09e: Rebase-only head onto current main with no UI change since the human-accepted a457e1a and a9109df; the error-state screenshots are embedded in the PR body.

@github-actions

Copy link
Copy Markdown
Contributor

Human judgment recorded

@chenmingwei23 marked the ux AI finding as false positive, not applicable, or explicitly accepted for 7fef09e50d31a9393da0f765620184048d7aa22f.

Rebase-only head onto current main with no UI change since the human-accepted a457e1a and a9109df; the error-state screenshots are embedded in the PR body.

This decision applies only to this commit. A new push requires a new judgment.

kirocrew-agent added 2 commits October 11, 2026 18:29
The chat side panel's Dashboard Overview now renders the root session's own
template-based Dynamic Dashboard, from the shared built-in catalog, with every
field folded from that session's own work ledger and crew log. Dispatched and
adopted sessions get no Overview.

- GET /api/chat/slots/{slot}/dashboard: the member dashboard read keyed by a
  root session's slot, owner-only, app-denied, same render and redaction.
- Instance store gains a slot-keyed directory beside the member one.
- Dashboard MCP tools accept a root session writing its own page; dispatched,
  adopted and slotless callers stay refused, and the crew webview publish
  route is unchanged.
- CrewDynamicDashboard takes a member or session target.
- Removes the status-tile HUD above the composer and the agent-published
  task-dashboard Overview, with their tests, i18n keys and prompt text.

Behind PREVIEW_DASHBOARD as before.

Refs #18022
A dashboard-only ErrorBoundary now wraps the lazy Overview Suspense, so a
failed chunk import stays contained and the Questions tab's unsent answer
draft survives instead of being unmounted with the whole panel.

A failed roster refresh is reported even when a cached DM row remains:
rosterFailed is roster.isError, so the error notice renders while the
cached DM dashboard stays mounted.
@chenmingwei23

Copy link
Copy Markdown
Contributor Author

/ai-review override ux a193a00: Rebase-only head onto current main with no UI change since the human-accepted a457e1a, a9109df and 7fef09e; the UX BLOCK is evidence-only (the two transient error-state notices command-center-roster-error and command-center-overview-error cannot be screenshotted into the committed stills). Error-state screenshots are embedded in the PR body / linked from branch screenshots/18022; the collector cannot read them.

@github-actions

Copy link
Copy Markdown
Contributor

Human judgment recorded

@chenmingwei23 marked the ux AI finding as false positive, not applicable, or explicitly accepted for a193a00991d23d1f3af918b3f7b7085fabb383c4.

Rebase-only head onto current main with no UI change since the human-accepted a457e1a, a9109df and 7fef09e; the UX BLOCK is evidence-only (the two transient error-state notices command-center-roster-error and command-center-overview-error cannot be screenshotted into the committed stills). Error-state screenshots are embedded in the PR body / linked from branch screenshots/18022; the collector cannot read them.

This decision applies only to this commit. A new push requires a new judgment.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Dynamic Dashboard in the chat side panel: root sessions only, replacing the agent-published task dashboard and the status-tile HUD

3 participants