@joekytc/dsh-swarm 0.1.2 → 0.2.0

package/skills/llm-wiki-skill/tests/fixtures/graph-interactive-multicomm/wiki/graph-data.json

@@ -0,0 +1,212 @@
+{
+  "meta": {
+    "build_date": "2026-04-23T12:00:00Z",
+    "wiki_title": "AI知识图谱Demo",
+    "total_nodes": 20,
+    "total_edges": 30,
+    "initial_view": ["n1","n2","n3","n4","n5","n6","n7","n8","n9","n10","n11","n12","n13","n14","n15","n16","n17","n18","n19","n20"],
+    "degraded": false
+  },
+  "nodes": [
+    {
+      "id": "n1", "label": "机器学习基础", "type": "topic", "community": "c1",
+      "content": "机器学习是人工智能的核心分支，通过数据驱动的方式让计算机自动发现模式并做出预测。它涵盖了监督学习、无监督学习和强化学习三大范式，是现代智能系统的基石。"
+    },
+    {
+      "id": "n2", "label": "深度学习", "type": "topic", "community": "c1",
+      "content": "深度学习是机器学习的子领域，利用多层神经网络自动提取数据的层次化特征表示。它在图像识别、语音处理和自然语言理解等任务上取得了突破性进展。"
+    },
+    {
+      "id": "n3", "label": "神经网络架构", "type": "entity", "community": "c1",
+      "content": "神经网络架构定义了神经元之间的连接方式和信息流动路径。常见的架构包括前馈网络、残差网络和U-Net，不同架构适用于不同类型的任务和数据特征。"
+    },
+    {
+      "id": "n4", "label": "卷积神经网络", "type": "entity", "community": "c1",
+      "content": "卷积神经网络通过局部感受野和权值共享机制高效处理网格状数据。它在图像分类、目标检测等视觉任务中表现出色，是计算机视觉领域最常用的网络结构。"
+    },
+    {
+      "id": "n5", "label": "循环神经网络", "type": "entity", "community": "c1",
+      "content": "循环神经网络通过隐藏状态在时间步之间传递信息，天然适合处理序列数据。LSTM和GRU等变体有效缓解了长程依赖中的梯度消失问题。"
+    },
+    {
+      "id": "n6", "label": "Transformer模型", "type": "entity", "community": "c1",
+      "content": "Transformer完全基于自注意力机制，摒弃了传统的循环和卷积结构。它通过并行计算大幅提升了训练效率，已成为NLP乃至多模态学习的主流架构。"
+    },
+    {
+      "id": "n7", "label": "注意力机制", "type": "entity", "community": "c1",
+      "content": "注意力机制让模型能够动态聚焦于输入中最相关的部分。多头注意力进一步扩展了这一能力，使模型能同时关注不同位置和不同表示子空间的信息。"
+    },
+    {
+      "id": "n8", "label": "自然语言处理", "type": "topic", "community": "c2",
+      "content": "自然语言处理研究如何让计算机理解、生成和转换人类语言。从早期的规则方法到现代预训练语言模型，NLP技术已广泛应用于翻译、摘要和对话系统。"
+    },
+    {
+      "id": "n9", "label": "词向量表示", "type": "entity", "community": "c2",
+      "content": "词向量将离散的词语映射为连续的稠密向量表示，使语义相近的词在向量空间中距离接近。Word2Vec和GloVe是两种经典的词向量生成方法。"
+    },
+    {
+      "id": "n10", "label": "语义分析", "type": "entity", "community": "c2",
+      "content": "语义分析旨在理解文本的深层含义，包括词义消歧、语义角色标注和情感分析等任务。它是实现机器真正理解自然语言的关键技术环节。"
+    },
+    {
+      "id": "n11", "label": "文本分类", "type": "entity", "community": "c2",
+      "content": "文本分类将文本文档自动归入预定义的类别中，广泛用于垃圾邮件过滤、新闻分类和意图识别。现代方法通常采用预训练模型加微调的策略。"
+    },
+    {
+      "id": "n12", "label": "语言模型", "type": "entity", "community": "c2",
+      "content": "语言模型学习自然语言的概率分布，能够预测序列中下一个词的概率。从N-gram到GPT系列，语言模型的能力和规模持续增长，推动了AI的快速发展。"
+    },
+    {
+      "id": "n13", "label": "计算机视觉", "type": "topic", "community": "c3",
+      "content": "计算机视觉让机器能够从图像和视频中提取有意义的信息。深度学习革命性地提升了视觉识别的准确率，使自动驾驶、医学影像分析等应用成为现实。"
+    },
+    {
+      "id": "n14", "label": "图像识别", "type": "entity", "community": "c3",
+      "content": "图像识别旨在将输入图像归类到已知类别中。从AlexNet到EfficientNet，卷积神经网络的演进不断提升ImageNet等基准上的分类精度。"
+    },
+    {
+      "id": "n15", "label": "目标检测", "type": "entity", "community": "c3",
+      "content": "目标检测不仅识别图像中的物体类别，还精确定位其位置。YOLO和Faster R-CNN等算法在速度和精度之间取得了良好的平衡。"
+    },
+    {
+      "id": "n16", "label": "图像分割", "type": "entity", "community": "c3",
+      "content": "图像分割将图像的每个像素分配到对应的语义类别，实现细粒度的场景理解。语义分割、实例分割和全景分割是三种主要的分割范式。"
+    },
+    {
+      "id": "n17", "label": "数据预处理", "type": "topic", "community": "c4",
+      "content": "数据预处理是机器学习流水线的第一步，涵盖缺失值填充、异常值处理和数据标准化等操作。高质量的预处理直接影响下游模型的训练效果和泛化能力。"
+    },
+    {
+      "id": "n18", "label": "特征工程", "type": "entity", "community": "c4",
+      "content": "特征工程从原始数据中构造有效的输入特征，是传统机器学习的核心技能。尽管深度学习减少了对人工特征的依赖，但在表格数据场景中仍然至关重要。"
+    },
+    {
+      "id": "n19", "label": "数据清洗", "type": "entity", "community": "c4",
+      "content": "数据清洗识别并修复数据集中的错误、不一致和冗余记录。常见问题包括重复条目、格式不统一和标签噪声，清洗质量直接决定模型可靠性。"
+    },
+    {
+      "id": "n20", "label": "特征选择", "type": "entity", "community": "c4",
+      "content": "特征选择从候选特征集中筛选出最相关的子集，降低维度灾难风险并提升模型可解释性。过滤法、包裹法和嵌入法是三种主流的特征选择策略。"
+    }
+  ],
+  "edges": [
+    {"id": "e1",  "from": "n1",  "to": "n2",  "type": "EXTRACTED", "weight": 0.95},
+    {"id": "e2",  "from": "n2",  "to": "n3",  "type": "EXTRACTED", "weight": 0.90},
+    {"id": "e3",  "from": "n2",  "to": "n4",  "type": "EXTRACTED", "weight": 0.88},
+    {"id": "e4",  "from": "n2",  "to": "n5",  "type": "EXTRACTED", "weight": 0.85},
+    {"id": "e5",  "from": "n2",  "to": "n6",  "type": "EXTRACTED", "weight": 0.92},
+    {"id": "e6",  "from": "n6",  "to": "n7",  "type": "EXTRACTED", "weight": 0.91},
+    {"id": "e7",  "from": "n3",  "to": "n4",  "type": "INFERRED", "weight": 0.70},
+    {"id": "e8",  "from": "n3",  "to": "n5",  "type": "INFERRED", "weight": 0.68},
+    {"id": "e9",  "from": "n8",  "to": "n9",  "type": "EXTRACTED", "weight": 0.93},
+    {"id": "e10", "from": "n8",  "to": "n10", "type": "EXTRACTED", "weight": 0.90},
+    {"id": "e11", "from": "n8",  "to": "n11", "type": "EXTRACTED", "weight": 0.87},
+    {"id": "e12", "from": "n8",  "to": "n12", "type": "EXTRACTED", "weight": 0.94},
+    {"id": "e13", "from": "n9",  "to": "n12", "type": "INFERRED", "weight": 0.72},
+    {"id": "e14", "from": "n10", "to": "n11", "type": "INFERRED", "weight": 0.65},
+    {"id": "e15", "from": "n13", "to": "n14", "type": "EXTRACTED", "weight": 0.91},
+    {"id": "e16", "from": "n13", "to": "n15", "type": "EXTRACTED", "weight": 0.89},
+    {"id": "e17", "from": "n13", "to": "n16", "type": "EXTRACTED", "weight": 0.86},
+    {"id": "e18", "from": "n14", "to": "n15", "type": "INFERRED", "weight": 0.73},
+    {"id": "e19", "from": "n15", "to": "n16", "type": "INFERRED", "weight": 0.67},
+    {"id": "e20", "from": "n17", "to": "n18", "type": "EXTRACTED", "weight": 0.90},
+    {"id": "e21", "from": "n17", "to": "n19", "type": "EXTRACTED", "weight": 0.92},
+    {"id": "e22", "from": "n17", "to": "n20", "type": "EXTRACTED", "weight": 0.88},
+    {"id": "e23", "from": "n18", "to": "n20", "type": "INFERRED", "weight": 0.75},
+    {"id": "e24", "from": "n19", "to": "n18", "type": "INFERRED", "weight": 0.60},
+    {"id": "e25", "from": "n6",  "to": "n12", "type": "EXTRACTED", "weight": 0.96},
+    {"id": "e26", "from": "n4",  "to": "n14", "type": "INFERRED", "weight": 0.80},
+    {"id": "e27", "from": "n1",  "to": "n17", "type": "INFERRED", "weight": 0.55},
+    {"id": "e28", "from": "n7",  "to": "n10", "type": "INFERRED", "weight": 0.62},
+    {"id": "e29", "from": "n5",  "to": "n12", "type": "INFERRED", "weight": 0.58},
+    {"id": "e30", "from": "n4",  "to": "n16", "type": "INFERRED", "weight": 0.64}
+  ],
+  "insights": {
+    "surprising_connections": [
+      {"from": "Transformer模型", "to": "语言模型", "weight": 0.96, "reason": "Transformer架构与语言模型的强关联，跨社区最高权重边"},
+      {"from": "卷积神经网络", "to": "图像识别", "weight": 0.80, "reason": "CNN从视觉向NLP迁移的技术纽带"}
+    ],
+    "isolated_nodes": [],
+    "bridge_nodes": [
+      {"id": "n6", "label": "Transformer模型", "community_count": 2, "reason": "连接深度学习与NLP两个社区的关键桥梁"},
+      {"id": "n4", "label": "卷积神经网络", "community_count": 2, "reason": "卷积网络在视觉和NLP之间的跨域应用"}
+    ],
+    "sparse_communities": [],
+    "meta": {"degraded": false}
+  },
+  "learning": {
+    "version": 1,
+    "entry": {
+      "recommended_start_node_id": "n1",
+      "recommended_start_reason": "community_hub",
+      "default_mode": "global"
+    },
+    "views": {
+      "path": {
+        "enabled": true,
+        "start_node_id": "n1",
+        "node_ids": ["n1","n2","n3","n4","n5","n6","n7"],
+        "degraded": false
+      },
+      "community": {
+        "enabled": true,
+        "community_id": "c1",
+        "label": "深度学习",
+        "node_ids": ["n1","n2","n3","n4","n5","n6","n7"],
+        "is_weak": false,
+        "degraded": false
+      },
+      "global": {
+        "enabled": true,
+        "node_ids": [],
+        "degraded": false
+      }
+    },
+    "communities": [
+      {
+        "id": "c1",
+        "label": "深度学习",
+        "node_count": 7,
+        "source_count": 4,
+        "is_primary": true,
+        "is_weak": false,
+        "internal_edge_weight": 5.82,
+        "recommended_start_node_id": "n1"
+      },
+      {
+        "id": "c2",
+        "label": "自然语言处理",
+        "node_count": 5,
+        "source_count": 3,
+        "is_primary": false,
+        "is_weak": false,
+        "internal_edge_weight": 4.37,
+        "recommended_start_node_id": "n8"
+      },
+      {
+        "id": "c3",
+        "label": "计算机视觉",
+        "node_count": 4,
+        "source_count": 2,
+        "is_primary": false,
+        "is_weak": false,
+        "internal_edge_weight": 3.21,
+        "recommended_start_node_id": "n13"
+      },
+      {
+        "id": "c4",
+        "label": "数据工程",
+        "node_count": 4,
+        "source_count": 2,
+        "is_primary": false,
+        "is_weak": false,
+        "internal_edge_weight": 2.85,
+        "recommended_start_node_id": "n17"
+      }
+    ],
+    "degraded": {
+      "path_to_community": false,
+      "community_to_global": false
+    }
+  }
+}

package/skills/llm-wiki-skill/tests/fixtures/lint-sample-wiki/index.md

@@ -0,0 +1,14 @@
+# 测试知识库 Index
+
+本文件是 lint-runner.sh 的测试夹具，故意包含若干已知问题用于验证脚本。
+
+## 实体页
+
+- [[Real]]
+- [[C++]]
+- [[Other]]
+- [[Ghost]]
+
+## 主题页
+
+- [[AIbasic]]

package/skills/llm-wiki-skill/tests/fixtures/lint-sample-wiki/wiki/entities/C++.md

@@ -0,0 +1,6 @@
+# C++
+
+用来验证正则特殊字符不误判。C++ 的文件名里有 `+`，是正则的量词。
+脚本必须用 grep -F 字面匹配，否则会误报。
+
+被 [[Other]] 引用，因此非孤立。

package/skills/llm-wiki-skill/tests/fixtures/lint-sample-wiki/wiki/entities/Lonely.md

@@ -0,0 +1,4 @@
+# Lonely
+
+这是一个孤立页面。没有任何其他页面引用它，index.md 里也没有列出它。
+脚本应该把它报告为"孤立"。

package/skills/llm-wiki-skill/tests/fixtures/lint-sample-wiki/wiki/entities/Other.md

@@ -0,0 +1,8 @@
+# Other
+
+这个页面包含各种链接场景，用于测试断链检测：
+
+- [[Real]]           — 有效引用
+- [[C++]]            — 特殊字符的有效引用
+- [[Missing]]        — 断链：Missing.md 不存在
+- [[Target|显示]]    — 别名语法断链：Target.md 不存在，期望脚本剥离"|显示"后报 Target

package/skills/llm-wiki-skill/tests/fixtures/lint-sample-wiki/wiki/entities/Real.md

@@ -0,0 +1,7 @@
+# Real
+
+这是一个被其他页面引用的实体页。
+
+## 相关页面
+
+- [[Other]]

package/skills/llm-wiki-skill/tests/fixtures/lint-sample-wiki/wiki/sources/2026-01-01-test-source.md

@@ -0,0 +1,15 @@
+---
+tags: [素材摘要]
+source_path: raw/articles/test-source.md
+images: 2
+image_paths:
+  - raw/assets/test-fig1.png
+  - raw/assets/test-fig2.jpg
+---
+
+# Test Source
+
+这是一个测试用的素材摘要页，故意不收录在 index.md 中。
+用于验证反向 index 一致性检查和图片资产一致性检查。
+
+引用 [[Real]]。

package/skills/llm-wiki-skill/tests/fixtures/lint-sample-wiki/wiki/sources/2026-02-01-inline-img.md

@@ -0,0 +1,12 @@
+---
+tags: [素材摘要]
+source_path: raw/articles/inline-img.md
+images: 1
+image_paths: ["raw/assets/inline-fig1.png"]
+---
+
+# Inline Image Test
+
+测试 inline array 格式的 image_paths 解析。
+
+引用 [[Real]]。

package/skills/llm-wiki-skill/tests/fixtures/lint-sample-wiki/wiki/topics/AIbasic.md

@@ -0,0 +1,3 @@
+# AIbasic
+
+这是一个主题页。引用 [[Real]]，验证脚本能扫到 topics/ 下的引用。

package/skills/llm-wiki-skill/tests/graph-aggregation-fallback-trial.regression-1.sh

@@ -0,0 +1,46 @@
+#!/bin/bash
+# Regression: no-new-dependency aggregation fallback browser performance trial
+
+set -euo pipefail
+
+REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+
+artifact_dir="${GRAPH_AGGREGATION_TRIAL_ARTIFACT_DIR:-}"
+if [ -z "$artifact_dir" ]; then
+  artifact_dir="$(mktemp -d)"
+fi
+mkdir -p "$artifact_dir"
+
+npm run build -w @llm-wiki/graph-engine > /dev/null 2>&1
+
+playwright_node_path="${GRAPH_TRIAL_PLAYWRIGHT_NODE_PATH:-}"
+if [ -z "$playwright_node_path" ]; then
+  playwright_node_path="$(
+    node -e 'const path=require("path"); try { console.log(path.dirname(path.dirname(require.resolve("playwright/package.json")))); } catch { process.exit(1); }' 2>/dev/null || true
+  )"
+fi
+if [ -z "$playwright_node_path" ] || [ ! -d "$playwright_node_path/playwright" ]; then
+  echo "ERROR: Playwright is not installed locally. Install Playwright for this checkout, or set GRAPH_TRIAL_PLAYWRIGHT_NODE_PATH." >&2
+  exit 1
+fi
+
+chrome_executable="${GRAPH_AGGREGATION_TRIAL_CHROME_EXECUTABLE:-}"
+if [ -z "$chrome_executable" ]; then
+  candidate="$(
+    NODE_PATH="$playwright_node_path" node -e 'const { chromium } = require("playwright"); console.log(chromium.executablePath())'
+  )"
+  if [ -x "$candidate" ]; then
+    chrome_executable="$candidate"
+  elif [ -x "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" ]; then
+    chrome_executable="/Applications/Google Chrome.app/Contents/MacOS/Google Chrome"
+  fi
+fi
+
+cd "$REPO_ROOT"
+NODE_PATH="$playwright_node_path" \
+GRAPH_AGGREGATION_TRIAL_ARTIFACT_DIR="$artifact_dir" \
+GRAPH_AGGREGATION_TRIAL_CHROME_EXECUTABLE="$chrome_executable" \
+node --import tsx tests/browser/graph-aggregation-fallback-trial.ts
+
+node tests/browser/validate-graph-trial-result.mjs "$artifact_dir/aggregation-fallback-trial-results.json"
+echo "PASS: aggregation fallback performance trial ($artifact_dir/aggregation-fallback-trial-results.json)"

package/skills/llm-wiki-skill/tests/graph-analysis-helper.regression-1.sh

@@ -0,0 +1,237 @@
+#!/bin/bash
+# Regression: graph analysis helper should compute weights, source omission, and insights deterministically
+
+set -euo pipefail
+
+REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+HELPER="$REPO_ROOT/scripts/graph-analysis.js"
+
+fail() {
+    echo "FAIL: $1" >&2
+    exit 1
+}
+
+assert_text_contains() {
+    local text="$1"
+    local expected="$2"
+
+    if ! printf '%s' "$text" | grep -F -- "$expected" > /dev/null; then
+        fail "Expected output to contain: $expected"
+    fi
+}
+
+test_helper_computes_weights_and_source_omission() {
+    local tmp_dir output
+    tmp_dir="$(mktemp -d)"
+
+    mkdir -p "$tmp_dir/wiki/entities"
+
+    cat > "$tmp_dir/wiki/entities/A.md" <<'EOF'
+---
+sources: ["same.pdf"]
+---
+
+# A
+
+[[B]]
+EOF
+
+    cat > "$tmp_dir/wiki/entities/B.md" <<'EOF'
+---
+sources: ["same.pdf"]
+---
+
+# B
+
+[[A]]
+EOF
+
+    cat > "$tmp_dir/wiki/entities/C.md" <<'EOF'
+# C
+
+[[A]]
+EOF
+
+    cat > "$tmp_dir/nodes.json" <<EOF
+[
+  {"id":"A","label":"A","type":"entity","source_path":"$tmp_dir/wiki/entities/A.md"},
+  {"id":"B","label":"B","type":"entity","source_path":"$tmp_dir/wiki/entities/B.md"},
+  {"id":"C","label":"C","type":"entity","source_path":"$tmp_dir/wiki/entities/C.md"}
+]
+EOF
+
+    cat > "$tmp_dir/edges.json" <<'EOF'
+[
+  {"id":"e1","from":"A","to":"B","type":"EXTRACTED"},
+  {"id":"e2","from":"B","to":"A","type":"EXTRACTED"},
+  {"id":"e3","from":"C","to":"A","type":"EXTRACTED"}
+]
+EOF
+
+    node "$HELPER" "$tmp_dir/nodes.json" "$tmp_dir/edges.json" "$tmp_dir/out.json" 0 500 250 1000 > /dev/null
+
+    output="$(jq -r '.edges[] | select(.from == "A" and .to == "B") | [.weight, .source_signal_available, .signals.source_overlap, .signals.type_affinity] | @tsv' "$tmp_dir/out.json")"
+    [ "$output" = $'0.667	true	1	1' ] || fail "Unexpected A-B edge metrics: $output"
+
+    output="$(jq -r '.edges[] | select(.from == "C" and .to == "A") | [.weight, .source_signal_available, (.signals.source_overlap // "null"), .signals.type_affinity] | @tsv' "$tmp_dir/out.json")"
+    [ "$output" = $'0.5	false	null	1' ] || fail "Expected missing sources to be omitted from denominator, got: $output"
+
+    rm -rf "$tmp_dir"
+}
+
+test_single_node_graph_louvain_returns_safely() {
+    local output
+    output="$(node - <<'NODE' "$HELPER"
+const helper = require(process.argv[2]);
+const result = helper.runLouvain(["Solo"], new Map());
+console.log(JSON.stringify(Object.fromEntries(result)));
+NODE
+)"
+    [ "$output" = '{"Solo":"Solo"}' ] || fail "Expected single-node Louvain to return solo assignment, got: $output"
+}
+
+test_parse_sources_yaml_multiline() {
+    local output
+    output="$(node - <<'NODE' "$HELPER"
+const helper = require(process.argv[2]);
+const fm = [
+  "title: Test",
+  "sources:",
+  "  - paper_a.pdf",
+  "  - paper_b.pdf",
+  "  - \"quoted.pdf\""
+].join("\n");
+const result = helper.parseSourcesFrontmatter(fm);
+console.log(JSON.stringify(result));
+NODE
+)"
+    local has_field parsed signal sources
+    has_field="$(printf '%s' "$output" | jq -r '.hasField')"
+    parsed="$(printf '%s' "$output" | jq -r '.parsed')"
+    signal="$(printf '%s' "$output" | jq -r '.signalAvailable')"
+    sources="$(printf '%s' "$output" | jq -r '.sources | join(",")')"
+    [ "$has_field" = "true" ] || fail "Expected hasField=true for YAML multiline, got: $has_field"
+    [ "$parsed" = "true" ] || fail "Expected parsed=true for YAML multiline, got: $parsed"
+    [ "$signal" = "true" ] || fail "Expected signalAvailable=true, got: $signal"
+    [ "$sources" = "paper_a.pdf,paper_b.pdf,quoted.pdf" ] || fail "Unexpected sources: $sources"
+}
+
+test_node_helper_bad_json_exits_with_error() {
+    local tmp_dir output
+    tmp_dir="$(mktemp -d)"
+
+    printf 'not json' > "$tmp_dir/bad.json"
+    printf '[]' > "$tmp_dir/edges.json"
+
+    if output="$(node "$HELPER" "$tmp_dir/bad.json" "$tmp_dir/edges.json" "$tmp_dir/out.json" 0 500 250 1000 2>&1)"; then
+        fail "graph-analysis.js should fail on invalid JSON input"
+    fi
+
+    assert_text_contains "$output" "Invalid JSON"
+    rm -rf "$tmp_dir"
+}
+
+test_helper_preserves_node_source_path() {
+    local tmp_dir output
+    tmp_dir="$(mktemp -d)"
+
+    mkdir -p "$tmp_dir/wiki/synthesis/sessions"
+
+    cat > "$tmp_dir/wiki/synthesis/sessions/A.md" <<'EOF'
+# A
+
+正文。
+EOF
+
+    cat > "$tmp_dir/nodes.json" <<EOF
+[
+  {"id":"A","label":"A","type":"synthesis","source_path":"$tmp_dir/wiki/synthesis/sessions/A.md"}
+]
+EOF
+
+    cat > "$tmp_dir/edges.json" <<'EOF'
+[]
+EOF
+
+    node "$HELPER" "$tmp_dir/nodes.json" "$tmp_dir/edges.json" "$tmp_dir/out.json" 0 500 250 1000 > /dev/null
+
+    output="$(jq -r '.nodes[] | select(.id == "A") | .source_path // ""' "$tmp_dir/out.json")"
+    [ "$output" = "$tmp_dir/wiki/synthesis/sessions/A.md" ] || fail "Expected source_path to survive graph analysis, got: $output"
+
+    rm -rf "$tmp_dir"
+}
+
+test_helper_reports_sparse_and_bridge_insights() {
+    local output
+
+    output="$(node - <<'NODE' "$HELPER"
+const helper = require(process.argv[2]);
+const nodesById = {
+  's-topic': { id: 's-topic', label: 'Sparse Topic' },
+  's1': { id: 's1', label: 'S1' },
+  's2': { id: 's2', label: 'S2' },
+  's3': { id: 's3', label: 'S3' },
+  's4': { id: 's4', label: 'S4' },
+  's5': { id: 's5', label: 'S5' },
+  'bridge': { id: 'bridge', label: 'Bridge' },
+  't2': { id: 't2', label: 'T2' },
+  'b1': { id: 'b1', label: 'B1' },
+  't3': { id: 't3', label: 'T3' },
+  'c1': { id: 'c1', label: 'C1' }
+};
+const edges = [
+  { from: 's-topic', to: 's1', type: 'EXTRACTED' },
+  { from: 's-topic', to: 's2', type: 'EXTRACTED' },
+  { from: 'bridge', to: 's-topic', type: 'EXTRACTED' },
+  { from: 'bridge', to: 't2', type: 'EXTRACTED' },
+  { from: 'bridge', to: 't3', type: 'EXTRACTED' },
+  { from: 't2', to: 'b1', type: 'EXTRACTED' },
+  { from: 't3', to: 'c1', type: 'EXTRACTED' }
+];
+const pairMetrics = new Map([
+  ['bridge\ts-topic', { weight: 0.8 }],
+  ['bridge\tt2', { weight: 0.8 }],
+  ['bridge\tt3', { weight: 0.8 }]
+]);
+const communityAssignments = new Map([
+  ['s-topic', 's-topic'],
+  ['s1', 's-topic'],
+  ['s2', 's-topic'],
+  ['s3', 's-topic'],
+  ['s4', 's-topic'],
+  ['s5', 's-topic'],
+  ['bridge', 'bridge'],
+  ['t2', 't2'],
+  ['b1', 't2'],
+  ['t3', 't3'],
+  ['c1', 't3']
+]);
+const insights = helper.buildInsights(nodesById, edges, pairMetrics, communityAssignments, {
+  nodeCount: Object.keys(nodesById).length,
+  edgeCount: edges.length,
+  maxInsightNodes: 250,
+  maxInsightEdges: 1000
+});
+console.log(JSON.stringify(insights));
+NODE
+)"
+
+    local bridge sparse
+    bridge="$(printf '%s' "$output" | jq -r '.bridge_nodes[]?.id')"
+    assert_text_contains "$bridge" "bridge"
+
+    sparse="$(printf '%s' "$output" | jq -r '.sparse_communities[]?.id')"
+    assert_text_contains "$sparse" "s-topic"
+}
+
+main() {
+    test_helper_computes_weights_and_source_omission
+    test_single_node_graph_louvain_returns_safely
+    test_parse_sources_yaml_multiline
+    test_node_helper_bad_json_exits_with_error
+    test_helper_preserves_node_source_path
+    test_helper_reports_sparse_and_bridge_insights
+    echo "PASS: graph analysis helper regression coverage"
+}
+
+main "$@"

package/skills/llm-wiki-skill/tests/graph-browser-large-performance.regression-1.sh

@@ -0,0 +1,39 @@
+#!/bin/bash
+# Regression: generated large graph browser performance measurement
+
+set -euo pipefail
+
+REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
+
+artifact_dir="${GRAPH_LARGE_PERF_ARTIFACT_DIR:-}"
+if [ -z "$artifact_dir" ]; then
+  artifact_dir="$(mktemp -d)"
+fi
+mkdir -p "$artifact_dir"
+
+npm run build -w @llm-wiki/graph-engine > /dev/null 2>&1
+
+playwright_node_path="$(
+  npx --yes -p playwright -c 'node -e "const path=require(\"path\"); console.log(path.dirname(process.env.PATH.split(\":\")[0]))"'
+)"
+
+chrome_executable="${GRAPH_LARGE_PERF_CHROME_EXECUTABLE:-}"
+if [ -z "$chrome_executable" ]; then
+  candidate="$(
+    NODE_PATH="$playwright_node_path" node -e 'const { chromium } = require("playwright"); console.log(chromium.executablePath())'
+  )"
+  if [ -x "$candidate" ]; then
+    chrome_executable="$candidate"
+  elif [ -x "/Applications/Google Chrome.app/Contents/MacOS/Google Chrome" ]; then
+    chrome_executable="/Applications/Google Chrome.app/Contents/MacOS/Google Chrome"
+  fi
+fi
+
+cd "$REPO_ROOT"
+NODE_PATH="$playwright_node_path" \
+GRAPH_LARGE_PERF_ARTIFACT_DIR="$artifact_dir" \
+GRAPH_LARGE_PERF_CHROME_EXECUTABLE="$chrome_executable" \
+node --import tsx tests/browser/graph-large-performance.ts
+
+test -f "$artifact_dir/large-graph-performance-results.json"
+echo "PASS: large graph performance measurement ($artifact_dir/large-graph-performance-results.json)"