feat(node-ui): PR6 — dashboard revamp + Context Graph terminology + bottom panel#583
Open
Jurij89 wants to merge 34 commits into
Open
feat(node-ui): PR6 — dashboard revamp + Context Graph terminology + bottom panel#583Jurij89 wants to merge 34 commits into
Jurij89 wants to merge 34 commits into
Conversation
Remove Gossip / Agent Runs / SPARQL tabs (keep Node Log + Transactions). Add a vertical drag-resize: layout store gains `bottomHeight` (clamp 120–600, default 200 = old fixed height, debounced persist to dkg-layout, clamp on load) mirroring leftWidth/rightWidth; App.tsx adds `useDragResizeV` (clientY twin of useDragResize) + a `.v10-resize-handle-v` above PanelBottom, hidden when collapsed; PanelBottom drives its height from the store. (Handle CSS lands with the dashboard styles commit.) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…phs list) Top row → My Context Graphs / Context Graph Size / Connected Agents (Knowledge Assets + Connected Peers stats dropped per plan). Context Graph Size shows TWO numbers (entities/KAs + triples), each with a WM/SWM/VM proportion bar using the shared --layer-* CSS tokens (not raw hex — fixes light theme) with a 1px track/separator so the light-theme WM segment stays legible (ui-lead). Removed Quick Actions + Recent Operations; kept Spending. "Projects" box → "My Context Graphs": filtered to the sidebar membership set, columns (name / Lock·Globe type icon / size / agents / curator·joined badge), 10-row internal scroll. One canonical "what is a Context Graph?" header affordance (not per-card tooltips — ux-lead); helper text always-visible in the sub slot; informational dim text uses --text-secondary (AA). Data: new shared `useHiddenContextGraphIds` (PanelLeft + MemoryStack refactored onto it — single source of truth so the dashboard count == sidebar count) and `useMyContextGraphs`. Per-CG `CgRow` is the single data probe (one useMemoryEntities + one listParticipants, reports up; parent dedups agents via canonicalAgentDid). `normalizeAccessPolicy` exported. Also carries the `.v10-resize-handle-v` CSS for the bottom panel. Per-CG SPARQL aggregation at PoC scale (MemoryStack pattern); bulk endpoint is a documented follow-up. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…st updates User-visible copy only across the create/join/share modals, the composer picker placeholder + attach hints (PanelRight), and the context-graph overview back-button title. Internal identifiers, store keys, CSS classes, tab ids and filenames are intentionally unchanged. Update tests to the new copy and remove the ui-compat block that guarded the now-removed dashboard "Import Memories" Quick Action (import still launches from the left sidebar). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
The PR6 dashboard revamp shipped two new shared hooks that back the sidebar↔dashboard membership-parity invariant but had no unit coverage (rubric: missing tests for high-risk user-facing behavior = blocker). Adds 11 tests: - useHiddenContextGraphIds: localStorage load, legacy-key back-compat, corrupt-payload tolerance, hide()/unhideAll() persistence, and the cross-instance custom-event sync that the parity guarantee depends on. - useMyContextGraphs: callerInvolved membership filter (identity- independent), hidden-id subtraction (the exact sidebar filter), case-insensitive curator-by-identity match, identityLoading lifecycle, and the hidden+non-member → empty-set edge. No source change (DashboardView reducer findings reported to team-lead for the in-flight sig rework). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
… state, badge AA 🔴 Codex: the onReport memo key compared only entity/triple totals + agents.length, so a WM→SWM/VM promotion or an agent swap left the dashboard cards stale until remount. Replace with a full `sig` (per-layer counts + sorted agent ids + flags) used as both the report effect dep and the parent dedupe key. 🟡 Codex: CgRow coerced a failed/in-flight participant probe to `[]`, so the Agents column / Connected Agents card showed a silent 0. Propagate `agentsLoading`/`agentsError`; the aggregate excludes errored CGs from the unique-agent union and marks the count partial. #86 (ui-lead): curator role badge used `--accent-green` text on `--bg-elevated` — ~1.9:1 in light theme (accent-green is theme-invariant, bg-elevated isn't). Invert to a solid green chip with fixed near-black text — AA in both themes and rightly emphasises the consequential role. Also adds the Connected Agents partial-caveat parity (ux-lead NB-2). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ale reports ui-lead: light `--layer-working` #a3a3a3 (~2.0:1 vs the white card) left the WM proportion-bar segment near-invisible. Not a hard WCAG blocker (the breakdown is redundantly available via the legend at --text-secondary 7.81:1 and the bar's title/aria-label), but nudged the light token to #8a8a8a (~3:1) so the segment has a visible edge. Dark unaffected. qa-lead: `reports` accumulated orphaned per-CG entries across hide/unhide. Added a guarded prune keyed on the membership set — memory hygiene (aggregate already iterates myCgs). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…rtial qa-lead round-2 🔵: the unified `loading`/`partial` flags made the Context Graph Size card flash "loading…" during an unrelated participant refetch (and vice-versa). Split the aggregate into `sizeLoading`/`agentsLoading` + `sizePartial`/`agentsPartial`; each card now reflects only its own data's state. No behavior change at rest; removes the cross-card transient. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ent copy 🔴 CgRow size cell gated on `mem.loading && agents===null`, so if /participants resolved before useMemoryEntities the row briefly showed a false "0 entities · 0 triples". Gate on `mem.loading` alone. 🟡 Bottom-panel height clamped to a static 600 only — a height persisted on a tall screen could squeeze the center pane out of view after reload on a shorter one. Add viewport-aware `maxBottomHeight()` (min center 240px) applied at drag time and at render. 🟡 `/participants` returns the access allow-list, not a live roster — relabel the Connected Agents helper + Agents column to "allow-listed with access" so the metric isn't over-claimed (it maps exactly to the user-spec'd "unique agent addresses across your context graphs"; 0 on fully-public graphs is correct for explicitly-granted access). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
… teardown, viewport clamp sync 🔴 The /participants allow-list can omit the curator (and is empty on public graphs), so the Agents count / Connected Agents stat under- counted. Fold the CG's canonical curator into the per-row agent set (they inherently have access); null loading/error semantics preserved. 🟡 useMyContextGraphs fetched identity once on mount — if it failed at startup or the agent later changed it permanently diverged from the sidebar, breaking the parity invariant. Re-fetch on context-graph-list change (same cadence as PanelLeft's reload-with-identity). 🔴/🟡 useDragResizeV cleanup only removed the mousedown listener; an unmount mid-drag leaked document mousemove/mouseup + body cursor/ user-select. Cleanup now calls onMouseUp() for a full teardown. 🔴 The bottom-height viewport clamp was render-only, leaving the store holding a stale large value so the first shrink-drag felt stuck and window resize didn't reclamp. Sync the clamp into the store on mount and on window resize so render + drag math agree. 🟡 Per-row useMemoryEntities SPARQL cost: unchanged — the explicit, plan/PR-documented PoC-scale decision (see PR thread). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…rk, Wallets&Spending) User round-1 feedback on the dashboard: - Remove the stray header "What is a Context Graph?" affordance; merge a concise CG definition into the My Context Graphs card under its sub (.v10-cg-defn, --text-secondary AA). Drop unused HelpCircle import. - Context Graph Size: drop the redundant combined top number; the two metrics ARE the headline now — big number (.v10-cg-size-big 24/700) + small unit label, each over its WM/SWM/VM bar; shared legend; tighter card. Copy clarifies totals are "across all your context graphs". - WM/SWM/VM legend + bar segments get concise per-layer hover descriptions (README + MemoryStackView desc). - Connected Agents sub → "Unique agents allow-listed and collaborating across your context graphs." - Spending → "Wallets and Spending": node wallets + TRAC/ETH balances (existing /api/wallets/balances, now wired through api-wrapper + mock) with loading/error/addresses-only states; a 3-row spending overview (Last 24h / 7d / 30d, publishes·TRAC) from the existing economics periods. No backend change. - My Context Graphs list: header cells now align to their column content (Type/Role centered, Agents right); role badge centered. - Widen dashboard max-width 1200 → 1600 so wide screens aren't a big empty gutter. tsc clean; full node-ui suite green (the 2 parallel-flake files pass 88/88 in isolation, as before). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…ements - CG definition: card footer with hairline divider, 11px (was inline 10px) - Size card: 26px numbers, 14/6 spacing, scoped tighter padding - Width cap 1600 → 1440; grid-2 → minmax(0,2.2fr) minmax(320px,1fr) - CG list: cols 40/52/76px, Role badge left-anchored (was centered) - Wallets & Spending: per-wallet TRAC+ETH, grouped formatting, 3-col spending mini-table with header - Section/stat-card lucide icons; gated 160ms fadeSlideIn mount Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Resolved-but-empty / error+no-wallets no longer fall through to a perpetual "Loading wallets…"; distinct loading vs "No node wallets found." vs error states - Spending mini-table TRAC column header is literal "TRAC" (economics totalTrac is TRAC-denominated regardless of node wallet symbol) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Wallets block rebuilt as one state machine on useFetch `loading`: loading → balances → addresses-only → error → "No node wallets found." (resolves 🔴 permanent "Loading wallets…") - 🟡 surface chain/RPC error as a stale caveat even when balances/ addresses are shown (mirrors agentsPartial); previously swallowed - 🔵 fmtTrac: empty/whitespace/nullish → "—" not misleading "0" (real 0.00 still renders 0) - Drop wb! non-null assertion in favour of (wb?.wallets ?? []) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- layout.maxBottomHeight: subtract fixed shell chrome (--header-h 44px) before clamping; bottom panel could grow ~44px too tall and push the center pane below CENTER_MIN_HEIGHT - DashboardView effectiveAgents: seed the locally-known curator even when listParticipants() fails, so a /participants outage no longer undercounts/zeros a CG (loading still null to preserve loading state) - DashboardView wallets: capture useFetch `error` (hard fetch rejection) — a rejected /api/wallets/balances now shows "Wallet balances unavailable." instead of false "No node wallets found." Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- useDragResizeV: bind via state-tracked callback ref so the bottom resize handle works after expanding (panel defaults collapsed, so the one-shot useEffect([]) bound to a null ref and never re-ran) - useMemoryEntities/queryLayer: propagate fetch failures instead of coercing to []; in mock/offline mode every CG was shown as a real 0/0 instead of lighting the existing size-unavailable/partial state - wallets stale caveat: also trigger on useFetch `wbError` (a later poll rejecting keeps stale `wb` but only set wbError) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- App.tsx: remove stray `return handleRef;` / `}` orphaned by the round-2 useDragResizeV refactor (top-level return — broke the clean Vite/tsc build; incremental tsc had masked it) - DashboardView aggregate: union r.agents unconditionally. The row already reports only known DIDs (curator fallback on participant- probe failure, [] while loading), so skipping the row on agentsError cancelled the fallback and dropped the curator to 0; agentsPartial still flags the undercount Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…parity - Rename "Connected Agents" card → "Collaborating Agents" (matches the allow-listed-and-collaborating sub-copy; drops the live-presence overclaim) — per product owner Codex round-6 (recurring mock/offline-parity root cause, fixed at source): - listParticipants now routes through api-wrapper with a mock (MOCK_PARTICIPANTS); mock CGs gain callerInvolved + curator. In mock/offline mode the dashboard now shows the mock context graphs and a real (deduped) agent count instead of 0/empty/partial - triples metric renders "—" (unknown) instead of a misleading exact 0 when the live /api/query failed and entities fell back to the cg.assetCount summary (no triple count exists in that summary) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- useMemoryEntities.fetchAll: Promise.allSettled per layer instead of
Promise.all. A single layer's timeout/500 no longer blanks WM/SWM/VM
for every consumer (ProjectView/MemoryStackView); partial results
are kept, and `error` is set only on total (all-three) failure —
which is the case the dashboard's assetCount fallback keys off
- Collaborating Agents sub copy: add a !hasCgs branch ("No context
graphs yet.") so a fresh empty node doesn't describe nonexistent
collaborators (mirrors the size card)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- useMemoryEntities: add a `partial` signal (some-but-not-all layers failed). error stays null in that case, but consumers can now tell counts are incomplete. New regression test for the one-layer-fails case (use-memory-entities-partial.test.ts) - DashboardView size fallback: use `cg.assetCount ?? cg.assets` (legacy-field compat, matching PanelLeft) so older daemon responses don't collapse a non-empty graph to 0 - DashboardView: report sizeError on mem.partial too (lights the "total is partial" caveat); per-row size cell shows the coarse summary count + "— triples" instead of a bare — when the live query failed, so rows aren't empty while the aggregate is populated Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- maxBottomHeight: stop flooring the viewport-derived max back up to BOTTOM_HEIGHT_MIN. On short viewports (e.g. 360px) that forced the panel to 120 and squeezed the center pane below CENTER_MIN_HEIGHT; now returns the true available space (→ 0/collapsed if needed) so the center keeps its minimum. The 120 floor is storage-sanity only, enforced on load - fetchCurrentAgent now routes through api-wrapper (mock-aware) + MOCK_AGENT_IDENTITY whose agentDid matches the mock curators, so mock/offline mode keeps curator-aware behavior (CURATOR badge, identity-fallback membership) in sync with the real daemon path. Test mocks it at the api-wrapper layer accordingly Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- useMemoryEntities: failed-vs-empty signalling is now opt-in (`signalErrors`, default false). queryLayer no longer throws by default, so ProjectView/MemoryStackView/AgentProfilePage keep their original empty-on-failure behavior in mock/offline mode. Only the dashboard opts in to get the unavailable/partial distinction it needs (Codex-sanctioned alternative) - Per-row participants now refresh on a 30s interval (matches the dashboard's other live data), so the agent count + aggregate don't go stale after join approvals/removals until a page reload - Per-row size cell: a `mem.partial` (some layers failed) state now renders "~N entities · ~N triples" + tooltip instead of silently presenting truncated counts as exact Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- useMyContextGraphs now hydrates the CG store itself (fetch on mount + 60s + node-event refresh) instead of depending on PanelLeft's side effect. A reload with a persisted collapsed sidebar no longer shows 0 context graphs (PanelLeft is unmounted while collapsed) - useMyContextGraphs resets identity + identityLoading before each refetch and clears identity on failure, so after an agent/node switch membership can't be computed against a stale DID (falls back to callerInvolved instead) - Per-row participants subscribe to join_approved/join_rejected/ project_synced node events (filtered to this CG), so collaborator counts reflect membership changes immediately instead of up to 30s later; the 30s poll remains the backstop Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- useMyContextGraphs: filter the node-event CG refresh to join_approved/join_rejected/project_synced (was firing on every SSE event incl. high-volume memory_graph_changed → continuous list+identity refetch with the dashboard open) - useMyContextGraphs: stop clearing identity before/on failed refetch (round-11 overcorrection). Keep the last good identity until a replacement fetch succeeds, so a transient /api/agent/identity blip no longer drops every graph for curator-DID-fallback daemons; a real agent switch still replaces it - PanelBottom: floor the expanded panel render height at the tab-strip height (--tab-h 36px) so the tabs + collapse control can't vanish on short viewports while bottomCollapsed is still false Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Gate the My Context Graphs count + list empty-state on identityLoading: older daemons (no callerInvolved) resolve membership only once /api/agent/identity returns, so an unresolved empty list now shows "loading…" / "Loading context graphs…" instead of a false "0 / No context graphs yet" flash on every load. Round-13 #2 (small-screen variant for the CG list) is intentionally deferred to a follow-up — a responsive redesign for a non-primary form factor is out of scope for this dashboard-revamp PR. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
…led refresh) listParticipants .catch now clears `agents` to null in addition to setting agentsError. Previously a failed *refresh* kept the last-good array, so effectiveAgents returned the stale full member list and the row/aggregate hid a real membership removal behind the "partial" caveat. null now routes through the error branch (curator-only fallback + parent marks total partial) instead of trusting stale data. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- useMemoryEntities: queryLayer no longer throws; returns
{ triples, ok }. `partial` is now computed UNCONDITIONALLY for
every caller (was dead unless a caller opted in, so
MemoryStackView/ProjectView showed truncated-but-exact-looking
counts on a single-layer 500). Only the *hard error* escalation
on total failure stays configurable via signalErrors. Triple data
still degrades to [] on failure, so no consumer behavior regresses
- layout.setBottomHeight: lower clamp bound is now
min(BOTTOM_HEIGHT_MIN, maxBottomHeight()) instead of a hard 120.
On short viewports the static floor snapped every drag back to 120
while render clamped to the smaller viewport max — handle inert,
store wrong. Normal viewports keep the 120 floor
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Size card: in pure-fallback (live entity probe unavailable, value is the cg.assetCount summary) the entities metric is now explicitly labelled "Knowledge Assets (summary)" with a tooltip, and its layer bar is hidden — instead of silently presenting a published-KA count under the all-layer "entities" unit (Codex) - Collaborating Agents: public context graphs now report unknown (null → excluded from row + aggregate) instead of a confidently wrong [curator] count — for public graphs the /participants allow-list is not authoritative membership (Codex) Round-16 #C (identity reuse after rejected refresh) intentionally NOT changed: reviewer guidance has oscillated (r11 clear → r12 keep → r16 clear). r12's keep-last-identity optimizes the common case (transient blip must not empty the dashboard); the rare node-switch + identity-fail stale edge needs a backend node-version signal to solve without regressing the common case — tracked as a follow-up. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Round-16's row-level fixes left the aggregate cards silently exact. Comprehensive propagation of unknown/approx state into CgReport: - New CgReport.agentsUnknown (public graph = unmeasurable membership, distinct from agentsError). Aggregate now marks agentsPartial for public graphs, and the Collaborating Agents value renders "~N" when partial — instead of a confident exact count that silently excluded public graphs as 0 - New CgReport.sizeFallback (entities.total is the cg.assetCount summary substitute). If ANY row used it, the Size card no longer sums summary + live counts under a clean "entities" label: shows "~N" + "entities / KA · approx." + tooltip and hides the (now unreliable) WM/SWM/VM bars. Pure all-fallback keeps the stronger "Knowledge Assets (summary)" treatment from round-16 Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Make the per-row cells agree with the aggregate cards (the unit/approx honesty fixes were card-level only): - Size row fallback labels the value "KA (summary)" + tooltip instead of "entities" (matches the card's "Knowledge Assets (summary)" treatment) when the live entity probe is unavailable - Agents row shows "~N" + tooltip on the curator-only degraded fallback (agentsError) instead of a bare authoritative number, matching the card's "~N" partial treatment Third yellow (per-CG fan-out cost) is the known architectural scale item — deferred and tracked as the backend batched-aggregates follow-up, not a round-1 code change. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- My Context Graphs: single merged description (drop footnote+divider); add curator/joined split bar + legend (RoleBar) - Better panel icons: Workflow (graph) / TrendingUp (size) / UsersRound (agents) - Size card: "Note: Entities become Knowledge Assets…" on its own line - CG list grid rebalanced: Name flexes widest, Size flexes (no hard min) so names stay readable on narrow widths; Agents stays numeric right-aligned with matching header (standard for numeric columns) - Glow/hover lift extended to the lower section panels - Subtitle gains chain name (client chainId→name map, no backend) - Wallets & Spending: chain name in subhead; balances as an aligned 3-col table (Wallet | TRAC | Gas(<token>)); spending mini-table grid fixed (fixed numeric columns so Publishes/TRAC sit under their own headers — was collapsing both right) - Spending column renamed "Publishes to VM"; db.getSpendingSummary now counts only TRAC-spending VM publishes so the count and TRAC are consistent (free SWM/local/testnet publishes excluded) + tooltip Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- layout/PanelBottom: derive SHELL_CHROME_HEIGHT and TAB_STRIP_H from the CSS custom properties (--header-h, --tab-h) at runtime instead of duplicating the px numbers in JS. New `readPxVar` helper; a CSS tweak can no longer silently break the clamp - useMyContextGraphs: flip `identityLoading=true` at the start of every identity refetch (was never going back to true after the initial load). RoleBar accepts a `refreshing` prop and shows a subtle dim (`.is-refreshing` opacity 0.55, with prefers-reduced- motion guard) + aria-busy while the refetch is in flight, so a node/agent switch no longer asserts the old split with confidence. Identity itself is still NOT cleared (preserves round-12's transient- blip protection) Two yellows skipped with on-thread rationale: (1) PanelLeft loader dedup is an out-of-scope refactor with real regression risk — tracked; (4) per-CG fan-out cost is the 4th restatement, already tracked as backend follow-up #96. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
|
|
||
| const agentsMounted = useRef(true); | ||
| const loadParticipants = useCallback(() => { | ||
| api.listParticipants(cg.id) |
There was a problem hiding this comment.
🔴 Bug: loadParticipants() can race with itself here. The mount load, 30s poll, and SSE-triggered refreshes all call the same promise chain, so a slower older response can arrive last and overwrite a newer allow-list. That will make the dashboard's agent counts jump backwards after a join/remove. Guard this with a request sequence/AbortController and ignore stale responses before calling setAgents/setAgentsError.
| // the agent identity (curator-DID fallback) arrives. Until then an | ||
| // empty list is "not known yet", not a real zero — show a loading | ||
| // state rather than a false "0 / No context graphs yet" flash (Codex). | ||
| const cgsResolving = identityLoading && myCgs.length === 0; |
There was a problem hiding this comment.
🟡 Issue: cgsResolving only tracks identityLoading, but the context-graph list is fetched independently inside useMyContextGraphs. On a cold load, fetchCurrentAgent() can settle before fetchContextGraphs(), which makes the dashboard briefly render 0 / No context graphs yet even when graphs exist. Expose a real list-loading flag from the hook (or delay the empty state until the first CG fetch settles).
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
--layer-*tokens, light-theme-safe, 1px separator/track for the low-contrast light WM segment). Removed Quick Actions + Recent Operations; kept Spending. "Projects" box → My Context Graphs: filtered to the sidebar membership set, columns (name / Lock·Globe type icon / size / agents / curator·joined badge), up to 10 rows then internal scroll. One canonical "what is a Context Graph?" header affordance; always-visible helper text; AA-safe dim text..v10-resize-handle-vmirroring the horizontal handle, persistedbottomHeight(clamp 120–600, default 200) in the layout store.Peer-count investigation
Confirmed not a bug — two different concepts: header/old-dashboard "peers" = unique peers (
new Set(remotePeer)); Network tab "connected" = total libp2p connections (one peer can hold several). Per direction, no UI change — the dashboard peers stat is removed anyway and the header is left as-is; reported here for the record.Data strategy / assumptions
MemoryStackViewpattern: oneuseMemoryEntities+ onelistParticipantsper CG (the list rows are the single probe; top cards consume their aggregate). Accepted at PoC scale per plan.useHiddenContextGraphIdshook + the samebelongsInMyProjectsSidebarpredicate.Recommended follow-ups
Test plan
🤖 Generated with Claude Code