[mvalancy]

Relationship (edge) type labels went missing after "Organize around connections", in the hierarchical layout and once the force layout settled. The renderer never draws a label over a card or another label: it drops a label it cannot place. Labels were dropped for two reasons. They were hidden below 60% zoom, and Organize and the hierarchical fit frame most graphs further out than that. And the layouts left no room along an edge for its label. This PR is the minimal fix for what the owner reported, "Organize around connections" and the force layout; the hierarchical layout stays as on dev. Label room in the hierarchical layout, icon chips, hub placement, and lines through cards or on top of each other are in #283.

What changes

• Labels show while card titles do. The label cutoff is now CARD_TITLES_HIDE_BELOW (0.34) instead of 0.6. It lives in the new lib/cardTitleZoom.ts, and the zoom-out levels of Zoom out: labelled levels from items to graphs to projects (#277) #279 (merged from dev) now take it from there, so cards and labels share one cutoff. Labels keep their 14 px screen size at any zoom.
• Small graphs are framed whole on desktops and tablets. For graphs of up to 60 items (WHOLE_FIT_MAX_CARDS), Organize frames the whole graph, down to that 0.34 zoom. This replaces the 0.65 floor that cut part of the graph off. Where even 0.34 is not enough, the camera holds 0.34, centred on the graph when it nearly fits and on the busiest item's neighbourhood otherwise. Graphs over 60 items keep the plain organizer, the 0.65 floor and the overview/LOD behaviour.
• Phones keep their floors. On a 390 px phone, two compact cards and a label side by side do not fit at any zoom where the cards' titles still fit whole. Below about 0.45 "System coordinator" breaks to "coordinato r", and at 0.5 the hub card is 131 px wide where topology-organization asks for 132. So after Organize a phone keeps dev's camera: the 0.65 floor, centred on the busiest item's neighbourhood, and Organize leaves label room for that zoom. Fitting a whole graph on a phone needs compact label chips (Layouts: compact label chips on phones, hub placement in hierarchical, wires through cards #283).
• Layouts leave room for every label. Each edge's visible length fits its measured label at the zoom the camera will frame it (labelClearance.ts). That zoom comes from the layouts at 100% and at the lowest zoom, plus at most two more (regula falsi). If wires still cross labels, a component spreads evenly by the smallest factor that frees them. A tall organized layout on a wide screen, or a wide one on a tall screen, is turned a quarter turn when that frames it closer without putting cards on cards or hiding a label. After the force layout settles, only cards the simulation placed itself are moved, only for labels with such a card at one end, and a move is kept only when it shows more of those labels; saved positions never move.
• Labels do not depend on the item ids. Real ids are random and break ties in the organizer and in the label order. A spread is now kept only where it was checked, and each finished layout (after packing and the quarter turn) is checked as a whole. The renderer places each label from where it was in the previous frame, so the spots picked while the camera moves can box a label out. In a graph with at most 60 relationships it now also tries a fresh placement, the one the layouts leave room for, when a label on screen found no spot (labelsMissed).
• Bounded work. All of this runs in the click handler, so it is limited to 60 cards, LAYOUT_CHECKS label checks and five layouts per click: about 100 ms for 60 cards and 90 relationships on a desktop.
• Labels are checked as turned boxes (oriented, separating axes) with every spot along the edge tried, so a turned label is no longer refused next to a card it does not touch. Each placement first keeps only the obstacles and wires that can reach it, and tests exact axis-aligned bounds before the turned-box test.
• Card titles never split a word. A compact card is at most 240 graph units wide with its title kept at 14 px, so below about 0.45 a long word no longer fits a line, and fitGraphText split it ("coordinato r", "Design mockup s"). Organize rests at 0.34 on a desktop, so the first press showed it. A word wider than its line is now cut and ends in "…" ("coordinat…", "Design mocku…") at every zoom. A path, URL or identifier wider than a line (anything but letters inside its punctuation, or camelCase) breaks inside instead, after a separator or before a capital, and shows in full while lines are left ("Update / packages/web/src/lib/ / labelClearance.ts"); "…" only ends the last line. Scripts written without spaces still break between characters. Unit tests wrap seven titles at widths 40 to 320 and fail if a line starts or ends inside a word (before the fix: "Sys" / "tem"), and check that paths, URLs and identifiers come back whole; the journey fails a stage when a card title line on screen splits a word.
• The "Show Mini-Map" button is now an obstacle for labels and the fit. On iPad a label sat under it.

Review fixes (since the first review)

1. Freeze on Organize (1.4 s for 60 cards, 2.9 s for 200): the search is skipped above 60 cards (plain layouts, as before) and bounded below that. Unit guard: Organize of 60 cards / 90 relationships finishes within a second (about 100 ms on a desktop); above 60 cards it equals the plain organizer at floors 0.34 and 0.65.
2. Unplaced graphs of any size got the 0.34 spread: the 60-item bound now applies whatever the floor, and to clearing labels after the force layout settles. (Since the second review, unplaced graphs open exactly as on dev; see below.)
3. Vacuous journey check: every checked stage must draw every relationship of the graph (drawn === edges) and have at least one edge with both cards on screen.
4. Hierarchical → force kept the whole-graph framing: restored, and since the third round the hierarchical layout and its framing are dev's.

CI on the first push also found:

• organize-labels waited for a filled Welcome graph that earlier CI tests had emptied. It now waits only for the graph to exist.
• item-symbols: two parallel labels beside a diagonal wire were flagged as overlapping because their axis-aligned bounds overlap, while their outlines are 34 px apart. The check now uses the turned outlines.
• topology-organization at 390 px: the hub card must be 132 px wide, and on a phone framed whole at 0.34 its title broke mid-word ("coordinato r"). Phones now keep their floors (above), and the test is dev's, unchanged.
• large-symbols: one zoom took 5.1 s against the 5 s limit. A browser profile of that zoom is dominated by layout reflow (getBoundingClientRect in SemanticGraphOverview and updateEdgePositions), not by label placement. It passed on the next push.

Local journey runs then showed the launch checklist on desktop losing one label in about half the runs (the ids decide the order). The unit test label room whatever the item ids covers five id orders that hid a label before (5 of 10 cases failed), and the renderer's fresh retry covers the order in which the camera's path placed labels. After both, the journey passed twice in a row locally.

Second review fixes (re-review at 63c8f28)

Rule for this round: what the owner's report does not need goes back to dev's behaviour. Each item has a test that failed before and passes now.

	Reverted to dev or fixed	Test (red before → green now)
Settled force layout threw a free card away when two saved cards hid their own label (4× then 16× its distance, then saved)	Fixed (clearFreeCards): only labels with a free card at one end count; steps and the spread are kept only when they show more of those labels, the spread moves only the free cards of those labels' groups and must show all their labels; a second pass moves nothing	one new card in a saved graph whose saved cards hide a label: the card stays put when its label shows (before: (600, 400) → (2170, 1200)); when its own label is hidden it moves less than 1.5 label gaps from its neighbour (before: 1746 units away) and does not move again over two more passes
Hierarchical rows no longer wrapped (hub with 100 dependents 25564 units wide instead of 988; with 20, 4828 instead of 988)	Reverted at every size (third round): the hierarchical layout, its framing and layoutAlgorithms.ts with its tests are dev's, byte for byte; the label-room rows and their tests are gone	dev's own tests; a hub with 100 dependents is dev's 988 × 3680 again
Unplaced graphs first opened spread for labels at zoom 0 ("Plan a launch" at 0.20)	Reverted: the first open uses dev's organizer and framing	Journey a graph of unplaced items opens no smaller than its titles and labels show: "Plan a launch" unplaced rests at ≥ 0.34 (with the old line: 0.197)
Paths, URLs and identifiers lost their tail at full size ("labelCle…")	Fixed (fitGraphText): such tokens break inside while lines are left; ordinary words still never split	shows a path, URL or identifier wider than the line in full… (failed on 63c8f28)
Fresh-spot label retry ran where it could not help (labels off screen or past the 32 cap)	Limited: runs only when a label on screen was tried and found no spot (labelsMissed)	a label is missed only when one on screen was tried and found no spot (the old condition counted an off-screen label and a capped one as missing)

Third round: the hierarchical layout is dev's at every size. The owner reported Organize and the force layout, and the label-room rows made a common shape, an epic with 20 tasks, 4828 units wide instead of 988 with the camera on the top rows. Label room in the hierarchical layout and the review's remaining notes are on #283.

Test

organize-labels is a blocking sequence in the pr/full profiles and the "toolchain, schematic, LOD, layout" CI shard, run on desktop and iPhone. It creates Welcome and a twelve-item launch checklist with fifteen relationships of mixed types (the graph that was reported), presses the real Organize control (on a phone, in the View sheet) and checks the view it rests on. It also opens "Plan a launch" with every item unplaced and checks the camera rests no lower than 0.34. At the camera the app rests on, edgeLabelAudit.ts reads every label from the DOM. It fails a stage if:

• not every relationship of the graph is drawn;
• an edge with both cards on screen has no label;
• a label's text is under 11 px;
• a label is under a card or on another label;
• a label is not hit at its centre (elementFromPoint) or lies outside its edge's visible gap;
• no edge has both cards on screen (except the launch checklist on iPhone; see Layouts: compact label chips on phones, hub placement in hierarchical, wires through cards #283);
• any card is off screen while the camera rests above the floor (0.34; on iPhone 0.65).

It measures once the "Layout saved" note has faded, at the view a person is left with.

Unit tests (labelClearance.test.ts) place labels the way the renderer does, over whole layouts of the example graphs and the hard shapes (a hub with 12 spokes, two hubs, a long chain, a dense cluster of 30, long labels) on three screen sizes: none is dropped. They include five id orders of the launch checklist and two connected hubs that hid a label before.

Red before (unchanged dev 3a629e2)

This is the first version of the journey run against the unchanged source. That version also measured the force layout after Organize ("force after") and graphs of unplaced items ("force (unplaced)"); the journey now checks Organize, the hierarchical layout and the first open of an unplaced graph. Each cell is the labels dropped out of the edges whose two cards were both on screen. No label was ever drawn over a card or another label: the symptom is always a missing label. "zoom" marks a camera below the old 0.6 cutoff, where every label was hidden.

Graph	Stage	Desktop	iPad	iPhone
Welcome	organized	3 of 3	3 of 3	3 of 3
Welcome	hierarchical	3 of 3	3 of 3	3 of 3
Welcome	force after	3 of 3	3 of 3	2 of 2
Welcome	force (unplaced)	1 of 3	1 of 3	1 of 1 (zoom 0.5)
Plan a launch	organized	7 of 8	4 of 7	0 of 1
Plan a launch	hierarchical	5 of 7	5 of 5	4 of 4
Plan a launch	force after	5 of 7	2 of 5	0 of 1
Plan a launch	force (unplaced)	10 of 11	11 of 11 (zoom 0.589)	2 of 2 (zoom 0.5)
Fix bugs	organized	3 of 3	3 of 3	2 of 2
Fix bugs	hierarchical	2 of 2	1 of 1	2 of 2
Fix bugs	force after	2 of 2	2 of 2	0 of 0
Fix bugs	force (unplaced)	3 of 3	3 of 3 (zoom 0.562)	0 of 0 (zoom 0.5)
Move house	organized	2 of 3	0 of 3	2 of 2
Move house	hierarchical	3 of 3	2 of 2	3 of 3
Move house	force after	2 of 2	0 of 1	1 of 1
Move house	force (unplaced)	1 of 3	1 of 3	2 of 2 (zoom 0.5)
Hub with 12 spokes	organized	4 of 11	0 of 6	0 of 1
Hub with 12 spokes	hierarchical	10 of 12	12 of 12	8 of 8
Hub with 12 spokes	force after	0 of 4	0 of 1	0 of 1
Hub with 12 spokes	force (unplaced)	12 of 12 (zoom 0.561)	7 of 12	8 of 8 (zoom 0.5)
Two connected hubs	organized	6 of 12	8 of 11	1 of 1
Two connected hubs	hierarchical	13 of 15	12 of 12	9 of 9
Two connected hubs	force after	7 of 12	1 of 5	0 of 1
Two connected hubs	force (unplaced)	15 of 15	15 of 15 (zoom 0.525)	0 of 0 (zoom 0.5)
Long chain	organized	11 of 11	11 of 11	2 of 2
Long chain	hierarchical	6 of 6	3 of 3	2 of 2
Long chain	force after	7 of 7	2 of 2	2 of 2
Long chain	force (unplaced)	11 of 11 (zoom 0.593)	9 of 11	4 of 4 (zoom 0.5)
Dense cluster of 30	organized	11 of 18	5 of 9	0 of 2
Dense cluster of 30	hierarchical	33 of 33	23 of 23	0 of 0
Dense cluster of 30	force after	7 of 14	3 of 8	0 of 1
Dense cluster of 30	force (unplaced)	47 of 47 (zoom 0.402)	47 of 47 (zoom 0.441)	2 of 2 (zoom 0.5)
Long relationship labels	organized	3 of 8	3 of 9	0 of 1
Long relationship labels	hierarchical	10 of 10	4 of 4	5 of 5
Long relationship labels	force after	0 of 5	0 of 4	0 of 0
Long relationship labels	force (unplaced)	8 of 10	10 of 10 (zoom 0.561)	2 of 2 (zoom 0.5)

Welcome is the graph a new user sees first. Before this change, all 3 of its labels were dropped after Organize on every device, and in the hierarchical layout too. On iPhone, Organize framed it at 0.65, with part of the graph off screen.

After (this branch after merging dev 25a6a5e, local run of organize-labels: 6 passed, desktop and iPhone)

Measured at the camera the app rests on after Organize. "Area" and "mean edge" are in graph units, centre to centre. "Edges in view" have both cards on screen, clear of the controls; none of them lost its label.

Graph	Device	Labels shown	Edges in view	Zoom	Cards on screen	Area	Mean edge
Welcome	Desktop	3/3	3	1.098	4/4	308k	409
Welcome	iPhone	1/3	1	0.65	2/4	344k	476
Launch checklist	Desktop	13/15	11	0.34	10/12	11332k	1012
Launch checklist	iPhone	4/15	1	0.65	2/12	3739k	554

"Plan a launch" with every item unplaced opens at 0.374 on desktop (12/12 cards on screen) and 0.5 on iPhone, as on dev. Like dev, that first view hides most of its labels (1 of 11 shown on desktop) until the person presses Organize.

Spread for Welcome went from area 301k and mean edge 300 (desktop, organized, before) to 308k and 409. On iPhone, Organize frames it at 0.65 as before, now with the label of the edge on screen shown (before: all dropped). Earlier local runs: item-wires 13 passed (topology-organization at 390 and 1440 px included), large-symbols passed, semantic-lod passed (earlier push), web unit tests 83 files / 825 tests passed, runner CLI tests 43 passed. After the third round (hierarchical as on dev): the affected unit suites (labelClearance, layoutAlgorithms, graphSymbolLayout, edgeLabelPlacement, topologyOrganization, edgeRouteGeometry, localItemLayout, graphFit) 177 tests passed, tsc clean, runner CLI tests 46 passed, playwright test --list lists the journey's 6 tests.

Before and after: Welcome, Organize

	Before	After
Desktop
iPhone

Launch checklist on desktop after Organize: 14 of 15 labels, none dropped among the 13 edges on screen.

Not in this PR: #283

• Compact label chips. On a phone, 14 px labels on every edge need more screen than a 12-item graph has, so phones keep their floors (above) and show the busiest part. Even dev's organizer frames the launch checklist whole only at 0.25 on iPhone, before any label room. An icon chip where a full label cannot fit would let the whole graph show.
• The hierarchical layout is dev's: no label room yet, and only DEPENDS_ON sets the rows, so a hub with mixed relationships (Launch plan) sits at the bottom.
• Lines through other cards, and lines on top of each other. For example, Press kit → Email campaign → Launch plan lie on one line after Organize.
• Card titles in Zoom out: labelled levels from items to graphs to projects (#277) #279 come back only above 0.38. A graph organized to rest between 0.34 and 0.38, after the person had zoomed out past 0.34, shows labels while titles are still hidden until the next zoom.

🤖 Generated with Claude Code

[github-actions]

🧪 Comprehensive Test Suite

• Unit suites (Node 22.18.0 & 24.x) — core, web, server, mcp-server: ✅ passed
• Installer & deploy config: ✅ passed

Full-stack smoke gate runs in the CI workflow.

[github-actions]

🧪 Comprehensive Test Suite

• Unit suites (Node 22.18.0 & 24.x) — core, web, server, mcp-server: ✅ passed
• Installer & deploy config: ✅ passed

Full-stack smoke gate runs in the CI workflow.

[github-actions]

🧪 Comprehensive Test Suite

• Unit suites (Node 22.18.0 & 24.x) — core, web, server, mcp-server: ✅ passed
• Installer & deploy config: ✅ passed

Full-stack smoke gate runs in the CI workflow.

[github-actions]

🧪 Comprehensive Test Suite

• Unit suites (Node 22.18.0 & 24.x) — core, web, server, mcp-server: ✅ passed
• Installer & deploy config: ✅ passed

Full-stack smoke gate runs in the CI workflow.

[github-actions]

🧪 Comprehensive Test Suite

• Unit suites (Node 22.18.0 & 24.x) — core, web, server, mcp-server: ✅ passed
• Installer & deploy config: ✅ passed

Full-stack smoke gate runs in the CI workflow.

[github-actions]

🧪 Comprehensive Test Suite

• Unit suites (Node 22.18.0 & 24.x) — core, web, server, mcp-server: ✅ passed
• Installer & deploy config: ✅ passed

Full-stack smoke gate runs in the CI workflow.

[mvalancy]

Independent review of 63c8f28 (whole PR against dev)

I ran the targeted unit suites and they all pass: labelClearance, graphSymbolLayout, topologyOrganization, layoutAlgorithms, edgeRouteGeometry, edgeLabelPlacement/Layout, graphFit, edgeGeometry, graphZoomContext, localItem*, organizationCamera. The numbers below come from throwaway probes run against the branch. Nothing was committed.

Freezes: bounded. Work is counted in checks, not time.

• Organize on 60 cards with 90, 300 or 600 relationships takes about 60 to 120 ms in Node. Hierarchical takes about 35 to 95 ms. The settle-time clearFreeCards (two passes) takes about 20 ms.
• computeCompactLayout on 2000 cards and 4000 edges takes 41 ms.
• The renderer's fresh retry runs only on label passes (every 15th low-alpha tick, zoom end and settle), not on every zoom tick. It costs at most 2× ITEM_LABEL_BUDGET.

Determinism: OK. There is no Math.random and no timing in these paths. Ids break ties as on dev, and the organizer and rows sort by id, so input order does not matter. One caveat: the whole-layout check only holds while LAYOUT_CHECKS lasts. Dense 60-card graphs run out of it (Organize of 60 cards × 200 edges: 97 of 200 labels hidden in a whole-graph placement), and there the id order still decides.

must-fix

1. clearFreeCards throws new cards far away (labelClearance.ts:683-703, InteractiveGraphVisualization.tsx:4154). Suppose any hidden label has both of its cards pinned, for example two saved cards too close together, which is the case this PR exists for. spreadForLabels(..., { moves: free }) then requires zero hidden labels over all edges. It can never get there, so it climbs to maxFactor and returns 4. Every free card moves to 4× its distance from the pinned centroid. The second pass does it again (16×), and persistAllPositions saves the result.

   • Probe: pinned A(0,0) and B(230,0), whose label is hidden; pinned C(0,400); free N(600,400), linked to C, with its own label already showing. After pass 1 N is at (2170,1200); after pass 2 it is at (8450,4400).
   • Real case: a saved graph plus one item an agent or the API added without a position.
   • Fix: judge a spread only on edges that have a free end, or on whether hidden labels decrease. Add a unit test for this mix of pinned and free cards.

2. Hierarchical rows with any cross-row edge never wrap, at every graph size (layoutAlgorithms.ts:305). The PR says "Graphs over 60 items keep the plain layouts", but computeCompactLayout itself changed. The over-60 test (labelClearance.test.ts:251-257) compares against the new function, so it cannot catch this.

   • Probe, desktop, 220×80 cards, a hub with N dependents:

     Items	dev	PR
     20	988×800	4828×224
     100	988×3680	25308×224
     200	988×7280	50908×224

     Random 2000 cards × 4000 edges comes out 92124 units wide.

   • Fix: keep dev's wrapping above WHOLE_FIT_MAX_CARDS, and pin that test to dev's output. Below 60 this is a design call: a 20-item "all depend on one epic" graph was framed near 0.9 on dev and is now held at 0.34. If that is intended, say so in the PR body.

should-fix

3. The first open of an unplaced graph is framed further out than on dev, often below the title cutoff (InteractiveGraphVisualization.tsx:2148, labelClearance.ts:430-432). The initial organization uses floor 0. When no label zoom fits, withFramedRoom returns the most spread layout, and the camera frames it below 0.34, where neither labels nor titles show.

   • Fit zoom, dev → PR, desktop unless noted:
     • Plan a launch: 0.370 → 0.201
     • Launch checklist: 0.343 → 0.262 (iPad 0.466 → 0.194)
     • Two connected hubs: 0.492 → 0.170
     • Hub with 12 spokes: 0.482 → 0.321
     • Long labels: 0.391 → 0.318
   • This hits imported or agent-created graphs (all items with no position).
   • Fix: on desktop and tablet, give it Organize's whole-fit floor and framing, or fall back to the plain organizer when nothing fits at floor 0.

4. fitGraphText drops the rest of a long token at scale 1, even with lines free (graphSymbolLayout.ts:82). Ordinary titles do not change at scale 1: I checked 18, and the only change was titles that already overflowed 3 lines, which now end on a whole word. But a token wider than the line (a path, URL, identifier or compound word) used to wrap and show whole, and now loses its tail:

   • Update packages/web/src/lib/labelClearance.ts → ["Update","packages/web/src/lib/labelCle…"] (dev showed all of it on 3 lines)
   • PaymentGatewayRetryReconciliationHandler → one line, PaymentGatewayRetryReconci…
   • Investigate https://…/works…, then crash: an ellipsis in the middle of a title

   Suggestion: ellipsize only on compact cards (scale < 1), or break such tokens at / . - _ or camelCase while lines remain.

   The new "no split words" test does fail on dev's fitGraphText (System coordinator at 40: ['Sys','tem']), so it is not vacuous.

5. The fresh label retry runs where it cannot help (InteractiveGraphVisualization.tsx:3798). It compares against every eligible candidate, but eligible ignores the viewport, and placeItemLabels stops at 32 labels. So with 33 to 60 relationships, or any edge off screen (any zoom-in), it runs on every label pass, never wins, and doubles the label work. Compare instead against in-viewport eligible candidates, capped at ITEM_LABEL_BUDGET.labels.

6. clearFreeCards.hiddenNow acts on labels it never tried (labelClearance.ts:625-633). On the call that uses up the work, hidden includes labels skipped for budget, and step() pushes their cards apart. spreadForLabels and turnToScreen already treat that case as "no answer"; hiddenNow should do the same.

7. Journey coverage gaps. The journey runs only on placed graphs, for Organize and hierarchical. It does not cover hierarchical → force, the unplaced first load (Complete Neo4j migration with DOM flash fixes and external access support #3) or the settle-time clearing (Complete startup automation and external access fixes #1). mode: 'unplaced' (organize-labels.spec.ts:46,58-65) is never used. The comment at unified.config.mjs:95 ("once the force layout settles … hard shapes") and the "force after / force (unplaced)" rows in the PR body describe stages the spec no longer has. The checks it does run are not vacuous.

Checked and fine

• Camera framing: desktop/tablet frame whole at 0.34 for ≤60 cards; phones keep 0.65 after Organize and 0.5 in hierarchical.
• Switching hierarchical back to force restores the earlier floor and framing.
• Organize and hierarchical pin cards, and clearFreeCards moves only cards with fx == null. Saved positions are not overwritten, apart from Complete startup automation and external access fixes #1.

nits (simplicity)

• Exported but used only inside their own module: LABEL_PADDING, labelWorldScale, labelSideOffset, spreadPoint, SPREAD_CHECKS, SPREAD_MAX_CARDS, FRAMED_STEPS (labelClearance.ts) and LABEL_SAMPLES (edgeRouteGeometry.ts). Likewise EXAMPLE_SHAPES, HARD_SHAPES and the *_SHAPE constants other than Welcome and Launch checklist (labelClearanceShapes.ts).
• SPREAD_CHECKS (4M) is the default work for spreadForLabels/clearFreeCards. Every production caller passes LAYOUT_CHECKS, so the default is a trap: a future caller that omits it gets a budget more than 10× larger. Default to LAYOUT_CHECKS.
• labelReport is test-only code in a production module.
• dragged is passed through into the placeItemLabels candidates.

Verdict: not ready. Fix #1 and #2 first. The rest of the design is sound: the work is bounded, the results are deterministic and the retry is not per frame.

🤖 Generated with Claude Code

[github-actions]

🧪 Comprehensive Test Suite

• Unit suites (Node 22.18.0 & 24.x) — core, web, server, mcp-server: ✅ passed
• Installer & deploy config: ✅ passed

Full-stack smoke gate runs in the CI workflow.

[github-actions]

🧪 Comprehensive Test Suite

• Unit suites (Node 22.18.0 & 24.x) — core, web, server, mcp-server: ✅ passed
• Installer & deploy config: ✅ passed

Full-stack smoke gate runs in the CI workflow.