@dreb/coding-agent 1.16.0 → 2.0.0

package/dist/core/search/poem.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"poem.js","sourceRoot":"","sources":["../../../src/core/search/poem.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAGH,OAAO,EAAE,YAAY,EAAsC,MAAM,YAAY,CAAC;AAmB9E,MAAM,cAAc,GAAqC;IACxD,UAAU,EAAE;QACX,IAAI,EAAE,CAAC;QACP,MAAM,EAAE,CAAC;QACT,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,CAAC;QACd,WAAW,EAAE,CAAC;QACd,UAAU,EAAE,CAAC;KACb;IACD,gBAAgB,EAAE;QACjB,IAAI,EAAE,CAAC;QACP,MAAM,EAAE,CAAC;QACT,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,CAAC;QACd,WAAW,EAAE,CAAC;QACd,UAAU,EAAE,CAAC;KACb;IACD,SAAS,EAAE;QACV,IAAI,EAAE,CAAC;QACP,MAAM,EAAE,CAAC;QACT,SAAS,EAAE,CAAC;QACZ,WAAW,EAAE,CAAC;QACd,WAAW,EAAE,CAAC;QACd,UAAU,EAAE,CAAC;KACb;CACD,CAAC;AAEF,+EAA+E;AAC/E,uDAAuD;AACvD,+EAA+E;AAE/E,MAAM,OAAO,GAAG,IAAI,CAAC;AAErB,+EAA+E;AAC/E,UAAU;AACV,+EAA+E;AAE/E;;GAEG;AACH,SAAS,SAAS,CAAC,UAAqC,EAAE,IAAY,EAAe;IACpF,IAAI,UAAU,CAAC,IAAI,IAAI,IAAI,EAAE,CAAC;QAC7B,OAAO,IAAI,GAAG,CAAC,UAAU,CAAC,IAAI,EAAE,CAAC,CAAC;IACnC,CAAC;IAED,MAAM,KAAK,GAAG,IAAI,GAAG,EAAU,CAAC;IAEhC,KAAK,MAAM,MAAM,IAAI,YAAY,EAAE,CAAC;QACnC,MAAM,KAAK,GAA4B,EAAE,CAAC;QAC1C,KAAK,MAAM,CAAC,EAAE,EAAE,MAAM,CAAC,IAAI,UAAU,EAAE,CAAC;YACvC,KAAK,CAAC,IAAI,CAAC,CAAC,EAAE,EAAE,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACvC,CAAC;QACD,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAClC,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;QAC3C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,EAAE,CAAC,EAAE,EAAE,CAAC;YAChC,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACxB,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC;AAAA,CACb;AAED,+EAA+E;AAC/E,+BAA+B;AAC/B,+EAA+E;AAE/E;;;;;;;;;;;;;;;;;;GAkBG;AACH,SAAS,oBAAoB,CAC5B,MAAsB,EACtB,OAAsB,EACtB,IAAY,EACiC;IAC7C,MAAM,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;IACxB,MAAM,MAAM,GAAG,IAAI,WAAW,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACtC,IAAI,WAAW,GAAG,CAAC,CAAC;IAEpB,mCAAmC;IACnC,MAAM,OAAO,GAAG,IAAI,KAAK,CAAS,CAAC,CAAC,CAAC;IACrC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE;QAAE,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IAE3C,KAAK,MAAM,MAAM,IAAI,YAAY,EAAE,CAAC;QACnC,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;QAC/B,IAAI,MAAM,KAAK,CAAC;YAAE,SAAS;QAC3B,WAAW,IAAI,MAAM,CAAC;QAEtB,8CAA8C;QAC9C,MAAM,MAAM,GAAG,OAAO,CAAC,KAAK,EAAE,CAAC;QAC/B,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAE3E,6CAA6C;QAC7C,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;QAE5B,oDAAoD;QACpD,iEAA+D;QAC/D,KAAK,IAAI,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,CAAC,EAAE,EAAE,EAAE,EAAE,CAAC;YAC/B,MAAM,CAAC,GAAG,MAAM,CAAC,EAAE,CAAC,CAAC;YACrB,MAAM,KAAK,GAAG,CAAC,GAAG,CAAC,CAAC;YACpB,KAAK,IAAI,EAAE,GAAG,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,CAAC,EAAE,EAAE,EAAE,EAAE,CAAC;gBACpC,MAAM,CAAC,KAAK,GAAG,MAAM,CAAC,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC;YACtC,CAAC;QACF,CAAC;IACF,CAAC;IAED,OAAO,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC;AAAA,CAC7B;AAED;;;;;;;;;GASG;AACH,SAAS,cAAc,CAAC,MAAmB,EAAE,CAAS,EAAE,WAAmB,EAAgB;IAC1F,MAAM,OAAO,GAAG,IAAI,YAAY,CAAC,CAAC,CAAC,CAAC;IACpC,MAAM,SAAS,GAAG,WAAW,GAAG,GAAG,CAAC;IAEpC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC5B,IAAI,MAAM,GAAG,CAAC,CAAC;QACf,IAAI,aAAa,GAAG,CAAC,CAAC;QACtB,IAAI,aAAa,GAAG,CAAC,CAAC;QACtB,MAAM,KAAK,GAAG,CAAC,GAAG,CAAC,CAAC;QAEpB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YAC5B,IAAI,CAAC,KAAK,CAAC;gBAAE,SAAS;YACtB,MAAM,KAAK,GAAG,MAAM,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;YAChC,MAAM,IAAI,KAAK,CAAC;YAChB,IAAI,KAAK,GAAG,SAAS;gBAAE,aAAa,EAAE,CAAC;YACvC,IAAI,KAAK,GAAG,SAAS;gBAAE,aAAa,EAAE,CAAC;QACxC,CAAC;QAED,MAAM,OAAO,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAC7D,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,GAAG,CAAC,aAAa,GAAG,OAAO,CAAC,CAAC,GAAG,CAAC,aAAa,GAAG,OAAO,CAAC,CAAC;IAChF,CAAC;IAED,OAAO,OAAO,CAAC;AAAA,CACf;AAED,+EAA+E;AAC/E,aAAa;AACb,+EAA+E;AAE/E;;;;;;;GAOG;AACH,MAAM,UAAU,QAAQ,CAAC,UAAqC,EAAE,SAAoB,EAAE,IAAI,GAAG,IAAI,EAAqB;IACrH,IAAI,UAAU,CAAC,IAAI,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAErC,uCAAqC;IACrC,MAAM,SAAS,GAAG,SAAS,CAAC,UAAU,EAAE,IAAI,CAAC,CAAC;IAE9C,wBAAwB;IACxB,MAAM,GAAG,GAAa,EAAE,CAAC;IACzB,MAAM,MAAM,GAAmB,EAAE,CAAC;IAElC,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;QAC5B,MAAM,CAAC,GAAG,UAAU,CAAC,GAAG,CAAC,EAAE,CAAE,CAAC;QAC9B,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACb,MAAM,CAAC,IAAI,CAAC;YACX,IAAI,EAAE,CAAC,CAAC,IAAI,IAAI,CAAC;YACjB,MAAM,EAAE,CAAC,CAAC,MAAM,IAAI,CAAC;YACrB,SAAS,EAAE,CAAC,CAAC,SAAS,IAAI,CAAC;YAC3B,WAAW,EAAE,CAAC,CAAC,WAAW,IAAI,CAAC;YAC/B,WAAW,EAAE,CAAC,CAAC,WAAW,IAAI,CAAC;YAC/B,UAAU,EAAE,CAAC,CAAC,UAAU,IAAI,CAAC;SAC7B,CAAC,CAAC;IACJ,CAAC;IAED,MAAM,CAAC,GAAG,GAAG,CAAC,MAAM,CAAC;IAErB,8BAA4B;IAC5B,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;QACb,OAAO,CAAC,EAAE,EAAE,EAAE,GAAG,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC,CAAC;IACrD,CAAC;IAED,sDAAsD;IACtD,MAAM,OAAO,GAAG,cAAc,CAAC,SAAS,CAAC,CAAC;IAC1C,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,GAAG,oBAAoB,CAAC,MAAM,EAAE,OAAO,EAAE,IAAI,CAAC,CAAC;IAE1E,4BAA4B;IAC5B,MAAM,OAAO,GAAG,cAAc,CAAC,MAAM,EAAE,CAAC,EAAE,WAAW,CAAC,CAAC;IAEvD,mDAAmD;IACnD,MAAM,KAAK,GAAG,IAAI,KAAK,CAAS,CAAC,CAAC,CAAC;IACnC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE;QAAE,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IACzC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC;IAE9C,OAAO,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,CAAC;QAChC,EAAE,EAAE,GAAG,CAAC,GAAG,CAAC;QACZ,MAAM,EAAE,MAAM,CAAC,GAAG,CAAC;QACnB,IAAI;KACJ,CAAC,CAAC,CAAC;AAAA,CACJ","sourcesContent":["/**\n * POEM — Pareto-Optimal Embedded Modeling, multi-metric ranking.\n *\n * Ranks search candidates across multiple relevance metrics without requiring\n * hand-tuned weights. Uses the TFPR (Top-Fraction Pareto Ranker) approach:\n * vectorized dominance matrix computation with column duplication for\n * query-type-dependent metric weighting.\n *\n * Algorithm:\n *  1. Prune: per-metric top-K → union of survivors\n *  2. Build objectives matrix with column duplication for query-type weighting\n *  3. For each objective column, sort candidates and accumulate pairwise\n *     dominance counts (duplicate columns contribute via weight multiplier)\n *  4. Compute fitness: meanDominance × (numDominating + ε) / (numSubmitting + ε)\n *  5. Sort by fitness, assign ranks\n *\n * References:\n *  - POEM paper: https://iopscience.iop.org/article/10.1088/2632-2153/ab891b\n *  - TFPR: https://github.com/merckgroup/aidd_tfpr\n *  - colourdle: https://github.com/aebrer/colourdle\n */\n\nimport type { QueryType } from \"./query-classifier.js\";\nimport { METRIC_NAMES, type MetricName, type MetricScores } from \"./types.js\";\n\n// ============================================================================\n// Public types\n// ============================================================================\n\nexport interface RankedCandidate {\n\tid: number;\n\tscores: MetricScores;\n\trank: number;\n}\n\n// ============================================================================\n// Column duplication config per query type\n// ============================================================================\n\n/** How many times each metric column appears in the objectives matrix. */\ntype ColumnWeights = Record<MetricName, number>;\n\nconst COLUMN_WEIGHTS: Record<QueryType, ColumnWeights> = {\n\tidentifier: {\n\t\tbm25: 2,\n\t\tcosine: 1,\n\t\tpathMatch: 1,\n\t\tsymbolMatch: 2,\n\t\timportGraph: 1,\n\t\tgitRecency: 1,\n\t},\n\tnatural_language: {\n\t\tbm25: 1,\n\t\tcosine: 2,\n\t\tpathMatch: 1,\n\t\tsymbolMatch: 1,\n\t\timportGraph: 1,\n\t\tgitRecency: 1,\n\t},\n\tpath_like: {\n\t\tbm25: 1,\n\t\tcosine: 1,\n\t\tpathMatch: 3,\n\t\tsymbolMatch: 1,\n\t\timportGraph: 1,\n\t\tgitRecency: 1,\n\t},\n};\n\n// ============================================================================\n// Fitness smoothing constant (avoids division by zero)\n// ============================================================================\n\nconst EPSILON = 0.05;\n\n// ============================================================================\n// Pruning\n// ============================================================================\n\n/**\n * Per-metric top-K pruning → union of surviving candidate IDs.\n */\nfunction pruneTopK(candidates: Map<number, MetricScores>, topK: number): Set<number> {\n\tif (candidates.size <= topK) {\n\t\treturn new Set(candidates.keys());\n\t}\n\n\tconst union = new Set<number>();\n\n\tfor (const metric of METRIC_NAMES) {\n\t\tconst pairs: Array<[number, number]> = [];\n\t\tfor (const [id, scores] of candidates) {\n\t\t\tpairs.push([id, scores[metric] ?? 0]);\n\t\t}\n\t\tpairs.sort((a, b) => b[1] - a[1]);\n\t\tconst limit = Math.min(topK, pairs.length);\n\t\tfor (let i = 0; i < limit; i++) {\n\t\t\tunion.add(pairs[i][0]);\n\t\t}\n\t}\n\n\treturn union;\n}\n\n// ============================================================================\n// Dominance matrix computation\n// ============================================================================\n\n/**\n * Build the dominance count matrix using the TFPR approach.\n *\n * For each objective (metric), sorts candidates and accumulates pairwise\n * dominance: if candidate i ranks above candidate j on an objective,\n * dominanceCounts[i][j] increases by the column weight.\n *\n * Duplicate columns (from column duplication) are handled by multiplying\n * the contribution by the weight rather than re-sorting — same result,\n * no redundant work.\n *\n * Uses Uint16Array to keep memory compact (max possible value per cell\n * is the sum of all weights, which is ≤ 10).\n *\n * @param scores  Dense array of MetricScores, indexed 0..n-1\n * @param weights Column weights from query type\n * @param topK    Only top-K per objective contribute to pairwise dominance\n * @returns Flat dominance count matrix [n × n] and the total weight sum\n */\nfunction buildDominanceCounts(\n\tscores: MetricScores[],\n\tweights: ColumnWeights,\n\ttopK: number,\n): [counts: Uint16Array, totalWeight: number] {\n\tconst n = scores.length;\n\tconst counts = new Uint16Array(n * n);\n\tlet totalWeight = 0;\n\n\t// Reusable index array for sorting\n\tconst indices = new Array<number>(n);\n\tfor (let i = 0; i < n; i++) indices[i] = i;\n\n\tfor (const metric of METRIC_NAMES) {\n\t\tconst weight = weights[metric];\n\t\tif (weight === 0) continue;\n\t\ttotalWeight += weight;\n\n\t\t// Sort candidates by this metric (descending)\n\t\tconst sorted = indices.slice();\n\t\tsorted.sort((a, b) => (scores[b][metric] ?? 0) - (scores[a][metric] ?? 0));\n\n\t\t// Only consider top-K for pairwise dominance\n\t\tconst k = Math.min(topK, n);\n\n\t\t// For each pair in the top-K where i ranks above j:\n\t\t// i dominates j on this objective → add weight to counts[i, j]\n\t\tfor (let ri = 0; ri < k; ri++) {\n\t\t\tconst i = sorted[ri];\n\t\t\tconst iBase = i * n;\n\t\t\tfor (let rj = ri + 1; rj < k; rj++) {\n\t\t\t\tcounts[iBase + sorted[rj]] += weight;\n\t\t\t}\n\t\t}\n\t}\n\n\treturn [counts, totalWeight];\n}\n\n/**\n * Compute fitness scores from the dominance count matrix.\n *\n * Fitness = meanDominance × (numDominating + ε) / (numSubmitting + ε)\n *\n * Where:\n *  - meanDominance = average normalized dominance across all other candidates\n *  - numDominating = count of candidates this one dominates (>50% of objectives)\n *  - numSubmitting = count of candidates this one fails to dominate (<50%)\n */\nfunction computeFitness(counts: Uint16Array, n: number, totalWeight: number): Float64Array {\n\tconst fitness = new Float64Array(n);\n\tconst threshold = totalWeight * 0.5;\n\n\tfor (let i = 0; i < n; i++) {\n\t\tlet sumDom = 0;\n\t\tlet numDominating = 0;\n\t\tlet numSubmitting = 0;\n\t\tconst iBase = i * n;\n\n\t\tfor (let j = 0; j < n; j++) {\n\t\t\tif (i === j) continue;\n\t\t\tconst count = counts[iBase + j];\n\t\t\tsumDom += count;\n\t\t\tif (count > threshold) numDominating++;\n\t\t\tif (count < threshold) numSubmitting++;\n\t\t}\n\n\t\tconst meanDom = n > 1 ? sumDom / ((n - 1) * totalWeight) : 0;\n\t\tfitness[i] = (meanDom * (numDominating + EPSILON)) / (numSubmitting + EPSILON);\n\t}\n\n\treturn fitness;\n}\n\n// ============================================================================\n// Public API\n// ============================================================================\n\n/**\n * Rank candidates using POEM / TFPR.\n *\n * @param candidates Map of candidateId → MetricScores (all values 0–1)\n * @param queryType  Query type for column duplication weighting\n * @param topK       Per-metric pruning limit (default: 1000)\n * @returns Candidates ordered best-first with assigned ranks (0 = best)\n */\nexport function poemRank(candidates: Map<number, MetricScores>, queryType: QueryType, topK = 1000): RankedCandidate[] {\n\tif (candidates.size === 0) return [];\n\n\t// 1. Prune: per-metric top-K → union\n\tconst surviving = pruneTopK(candidates, topK);\n\n\t// 2. Build dense arrays\n\tconst ids: number[] = [];\n\tconst scores: MetricScores[] = [];\n\n\tfor (const id of surviving) {\n\t\tconst s = candidates.get(id)!;\n\t\tids.push(id);\n\t\tscores.push({\n\t\t\tbm25: s.bm25 ?? 0,\n\t\t\tcosine: s.cosine ?? 0,\n\t\t\tpathMatch: s.pathMatch ?? 0,\n\t\t\tsymbolMatch: s.symbolMatch ?? 0,\n\t\t\timportGraph: s.importGraph ?? 0,\n\t\t\tgitRecency: s.gitRecency ?? 0,\n\t\t});\n\t}\n\n\tconst n = ids.length;\n\n\t// Single candidate → rank 0\n\tif (n === 1) {\n\t\treturn [{ id: ids[0], scores: scores[0], rank: 0 }];\n\t}\n\n\t// 3. Compute dominance matrix with column duplication\n\tconst weights = COLUMN_WEIGHTS[queryType];\n\tconst [counts, totalWeight] = buildDominanceCounts(scores, weights, topK);\n\n\t// 4. Compute fitness scores\n\tconst fitness = computeFitness(counts, n, totalWeight);\n\n\t// 5. Sort by fitness (descending) and assign ranks\n\tconst order = new Array<number>(n);\n\tfor (let i = 0; i < n; i++) order[i] = i;\n\torder.sort((a, b) => fitness[b] - fitness[a]);\n\n\treturn order.map((idx, rank) => ({\n\t\tid: ids[idx],\n\t\tscores: scores[idx],\n\t\trank,\n\t}));\n}\n"]}

package/dist/core/search/query-classifier.d.ts

@@ -1,17 +0,0 @@
-/**
- * Classify search queries into types for POEM column weighting.
- *
- * Query types affect how metric columns are duplicated during ranking:
- * - identifier: emphasise BM25 and symbol-match scores
- * - path_like: emphasise path-match scores
- * - natural_language: emphasise cosine similarity scores
- */
-export type QueryType = "identifier" | "natural_language" | "path_like";
-/**
- * Classify a search query to guide POEM column weighting.
- *
- * @param query Raw user query string
- * @returns The detected query type
- */
-export declare function classifyQuery(query: string): QueryType;
-//# sourceMappingURL=query-classifier.d.ts.map

package/dist/core/search/query-classifier.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"query-classifier.d.ts","sourceRoot":"","sources":["../../../src/core/search/query-classifier.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAEH,MAAM,MAAM,SAAS,GAAG,YAAY,GAAG,kBAAkB,GAAG,WAAW,CAAC;AAoBxE;;;;;GAKG;AACH,wBAAgB,aAAa,CAAC,KAAK,EAAE,MAAM,GAAG,SAAS,CAyBtD","sourcesContent":["/**\n * Classify search queries into types for POEM column weighting.\n *\n * Query types affect how metric columns are duplicated during ranking:\n * - identifier: emphasise BM25 and symbol-match scores\n * - path_like: emphasise path-match scores\n * - natural_language: emphasise cosine similarity scores\n */\n\nexport type QueryType = \"identifier\" | \"natural_language\" | \"path_like\";\n\n/** Matches camelCase or PascalCase boundaries (lowercase→uppercase). */\nconst CAMEL_RE = /[a-z][A-Z]/;\n\n/** Matches snake_case — word chars around an underscore. */\nconst SNAKE_RE = /\\w+_\\w+/;\n\n/** SCREAMING_SNAKE_CASE — two or more uppercase-letter groups joined by underscores. */\nconst SCREAMING_SNAKE_RE = /^[A-Z][A-Z0-9]*(?:_[A-Z0-9]+)+$/;\n\n/** Path separators. */\nconst PATH_SEP_RE = /[/\\\\]/;\n\n/** Dotted path like `foo.bar.baz` (3+ segments). */\nconst DOTTED_PATH_RE = /^\\w+\\.\\w+\\.\\w+/;\n\n/** File extension pattern — ends with `.ext` where ext is 1-5 alphanumeric chars. */\nconst FILE_EXT_RE = /\\.\\w{1,5}$/;\n\n/**\n * Classify a search query to guide POEM column weighting.\n *\n * @param query Raw user query string\n * @returns The detected query type\n */\nexport function classifyQuery(query: string): QueryType {\n\tconst trimmed = query.trim();\n\tif (trimmed.length === 0) return \"natural_language\";\n\n\t// --- path_like ---\n\tif (PATH_SEP_RE.test(trimmed)) return \"path_like\";\n\tif (DOTTED_PATH_RE.test(trimmed)) return \"path_like\";\n\t// File extension at end of a single token (e.g. \"config.yaml\", \"auth.ts\")\n\tconst words = trimmed.split(/\\s+/);\n\tif (words.length === 1 && FILE_EXT_RE.test(trimmed) && /\\./.test(trimmed)) {\n\t\treturn \"path_like\";\n\t}\n\n\t// --- identifier ---\n\t// Single token or short (≤3 words) with code-style naming\n\tif (words.length === 1) return \"identifier\";\n\tif (words.length <= 3) {\n\t\t// If any word looks like a code identifier, classify as identifier\n\t\tif (words.some((w) => CAMEL_RE.test(w) || SNAKE_RE.test(w) || SCREAMING_SNAKE_RE.test(w))) {\n\t\t\treturn \"identifier\";\n\t\t}\n\t}\n\n\t// --- natural_language ---\n\treturn \"natural_language\";\n}\n"]}

package/dist/core/search/query-classifier.js

@@ -1,54 +0,0 @@
-/**
- * Classify search queries into types for POEM column weighting.
- *
- * Query types affect how metric columns are duplicated during ranking:
- * - identifier: emphasise BM25 and symbol-match scores
- * - path_like: emphasise path-match scores
- * - natural_language: emphasise cosine similarity scores
- */
-/** Matches camelCase or PascalCase boundaries (lowercase→uppercase). */
-const CAMEL_RE = /[a-z][A-Z]/;
-/** Matches snake_case — word chars around an underscore. */
-const SNAKE_RE = /\w+_\w+/;
-/** SCREAMING_SNAKE_CASE — two or more uppercase-letter groups joined by underscores. */
-const SCREAMING_SNAKE_RE = /^[A-Z][A-Z0-9]*(?:_[A-Z0-9]+)+$/;
-/** Path separators. */
-const PATH_SEP_RE = /[/\\]/;
-/** Dotted path like `foo.bar.baz` (3+ segments). */
-const DOTTED_PATH_RE = /^\w+\.\w+\.\w+/;
-/** File extension pattern — ends with `.ext` where ext is 1-5 alphanumeric chars. */
-const FILE_EXT_RE = /\.\w{1,5}$/;
-/**
- * Classify a search query to guide POEM column weighting.
- *
- * @param query Raw user query string
- * @returns The detected query type
- */
-export function classifyQuery(query) {
-    const trimmed = query.trim();
-    if (trimmed.length === 0)
-        return "natural_language";
-    // --- path_like ---
-    if (PATH_SEP_RE.test(trimmed))
-        return "path_like";
-    if (DOTTED_PATH_RE.test(trimmed))
-        return "path_like";
-    // File extension at end of a single token (e.g. "config.yaml", "auth.ts")
-    const words = trimmed.split(/\s+/);
-    if (words.length === 1 && FILE_EXT_RE.test(trimmed) && /\./.test(trimmed)) {
-        return "path_like";
-    }
-    // --- identifier ---
-    // Single token or short (≤3 words) with code-style naming
-    if (words.length === 1)
-        return "identifier";
-    if (words.length <= 3) {
-        // If any word looks like a code identifier, classify as identifier
-        if (words.some((w) => CAMEL_RE.test(w) || SNAKE_RE.test(w) || SCREAMING_SNAKE_RE.test(w))) {
-            return "identifier";
-        }
-    }
-    // --- natural_language ---
-    return "natural_language";
-}
-//# sourceMappingURL=query-classifier.js.map

package/dist/core/search/query-classifier.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"query-classifier.js","sourceRoot":"","sources":["../../../src/core/search/query-classifier.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAIH,0EAAwE;AACxE,MAAM,QAAQ,GAAG,YAAY,CAAC;AAE9B,8DAA4D;AAC5D,MAAM,QAAQ,GAAG,SAAS,CAAC;AAE3B,0FAAwF;AACxF,MAAM,kBAAkB,GAAG,iCAAiC,CAAC;AAE7D,uBAAuB;AACvB,MAAM,WAAW,GAAG,OAAO,CAAC;AAE5B,oDAAoD;AACpD,MAAM,cAAc,GAAG,gBAAgB,CAAC;AAExC,uFAAqF;AACrF,MAAM,WAAW,GAAG,YAAY,CAAC;AAEjC;;;;;GAKG;AACH,MAAM,UAAU,aAAa,CAAC,KAAa,EAAa;IACvD,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;IAC7B,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,kBAAkB,CAAC;IAEpD,oBAAoB;IACpB,IAAI,WAAW,CAAC,IAAI,CAAC,OAAO,CAAC;QAAE,OAAO,WAAW,CAAC;IAClD,IAAI,cAAc,CAAC,IAAI,CAAC,OAAO,CAAC;QAAE,OAAO,WAAW,CAAC;IACrD,0EAA0E;IAC1E,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IACnC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,IAAI,WAAW,CAAC,IAAI,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC;QAC3E,OAAO,WAAW,CAAC;IACpB,CAAC;IAED,qBAAqB;IACrB,4DAA0D;IAC1D,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,YAAY,CAAC;IAC5C,IAAI,KAAK,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;QACvB,mEAAmE;QACnE,IAAI,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,kBAAkB,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;YAC3F,OAAO,YAAY,CAAC;QACrB,CAAC;IACF,CAAC;IAED,2BAA2B;IAC3B,OAAO,kBAAkB,CAAC;AAAA,CAC1B","sourcesContent":["/**\n * Classify search queries into types for POEM column weighting.\n *\n * Query types affect how metric columns are duplicated during ranking:\n * - identifier: emphasise BM25 and symbol-match scores\n * - path_like: emphasise path-match scores\n * - natural_language: emphasise cosine similarity scores\n */\n\nexport type QueryType = \"identifier\" | \"natural_language\" | \"path_like\";\n\n/** Matches camelCase or PascalCase boundaries (lowercase→uppercase). */\nconst CAMEL_RE = /[a-z][A-Z]/;\n\n/** Matches snake_case — word chars around an underscore. */\nconst SNAKE_RE = /\\w+_\\w+/;\n\n/** SCREAMING_SNAKE_CASE — two or more uppercase-letter groups joined by underscores. */\nconst SCREAMING_SNAKE_RE = /^[A-Z][A-Z0-9]*(?:_[A-Z0-9]+)+$/;\n\n/** Path separators. */\nconst PATH_SEP_RE = /[/\\\\]/;\n\n/** Dotted path like `foo.bar.baz` (3+ segments). */\nconst DOTTED_PATH_RE = /^\\w+\\.\\w+\\.\\w+/;\n\n/** File extension pattern — ends with `.ext` where ext is 1-5 alphanumeric chars. */\nconst FILE_EXT_RE = /\\.\\w{1,5}$/;\n\n/**\n * Classify a search query to guide POEM column weighting.\n *\n * @param query Raw user query string\n * @returns The detected query type\n */\nexport function classifyQuery(query: string): QueryType {\n\tconst trimmed = query.trim();\n\tif (trimmed.length === 0) return \"natural_language\";\n\n\t// --- path_like ---\n\tif (PATH_SEP_RE.test(trimmed)) return \"path_like\";\n\tif (DOTTED_PATH_RE.test(trimmed)) return \"path_like\";\n\t// File extension at end of a single token (e.g. \"config.yaml\", \"auth.ts\")\n\tconst words = trimmed.split(/\\s+/);\n\tif (words.length === 1 && FILE_EXT_RE.test(trimmed) && /\\./.test(trimmed)) {\n\t\treturn \"path_like\";\n\t}\n\n\t// --- identifier ---\n\t// Single token or short (≤3 words) with code-style naming\n\tif (words.length === 1) return \"identifier\";\n\tif (words.length <= 3) {\n\t\t// If any word looks like a code identifier, classify as identifier\n\t\tif (words.some((w) => CAMEL_RE.test(w) || SNAKE_RE.test(w) || SCREAMING_SNAKE_RE.test(w))) {\n\t\t\treturn \"identifier\";\n\t\t}\n\t}\n\n\t// --- natural_language ---\n\treturn \"natural_language\";\n}\n"]}

package/dist/core/search/scanner.d.ts

@@ -1,30 +0,0 @@
-/**
- * File scanner for the semantic search subsystem.
- *
- * Discovers project files for indexing by walking the directory tree,
- * respecting .gitignore rules, and classifying files by type.
- */
-import type { FileType } from "./types.js";
-/** A file discovered by the scanner, ready for indexing. */
-export interface ScannedFile {
-    /** Path relative to the project root (posix separators). */
-    filePath: string;
-    /** Detected file type. */
-    fileType: FileType;
-    /** File modification time in milliseconds since epoch. */
-    mtime: number;
-}
-/**
- * Detect the {@link FileType} for a file path based on its extension.
- * Returns `null` for unrecognized extensions or files without an extension.
- */
-export declare function detectFileType(filePath: string): FileType | null;
-/**
- * Scan a project directory and return all indexable files.
- *
- * Walks the tree rooted at {@link projectRoot}, respects `.gitignore` rules,
- * skips binary / oversized files, and optionally includes memory files from
- * a global memory directory.
- */
-export declare function scanProject(projectRoot: string, globalMemoryDir?: string): Promise<ScannedFile[]>;
-//# sourceMappingURL=scanner.d.ts.map

package/dist/core/search/scanner.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"scanner.d.ts","sourceRoot":"","sources":["../../../src/core/search/scanner.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAMH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AAM3C,4DAA4D;AAC5D,MAAM,WAAW,WAAW;IAC3B,4DAA4D;IAC5D,QAAQ,EAAE,MAAM,CAAC;IACjB,0BAA0B;IAC1B,QAAQ,EAAE,QAAQ,CAAC;IACnB,0DAA0D;IAC1D,KAAK,EAAE,MAAM,CAAC;CACd;AAiED;;;GAGG;AACH,wBAAgB,cAAc,CAAC,QAAQ,EAAE,MAAM,GAAG,QAAQ,GAAG,IAAI,CAIhE;AAED;;;;;;GAMG;AACH,wBAAsB,WAAW,CAAC,WAAW,EAAE,MAAM,EAAE,eAAe,CAAC,EAAE,MAAM,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC,CAuBvG","sourcesContent":["/**\n * File scanner for the semantic search subsystem.\n *\n * Discovers project files for indexing by walking the directory tree,\n * respecting .gitignore rules, and classifying files by type.\n */\n\nimport { existsSync, readdirSync, readFileSync, type Stats, statSync } from \"node:fs\";\nimport { homedir } from \"node:os\";\nimport { extname, isAbsolute, join, relative, sep } from \"node:path\";\nimport ignore from \"ignore\";\nimport type { FileType } from \"./types.js\";\n\n// ============================================================================\n// Public types\n// ============================================================================\n\n/** A file discovered by the scanner, ready for indexing. */\nexport interface ScannedFile {\n\t/** Path relative to the project root (posix separators). */\n\tfilePath: string;\n\t/** Detected file type. */\n\tfileType: FileType;\n\t/** File modification time in milliseconds since epoch. */\n\tmtime: number;\n}\n\n// ============================================================================\n// Constants\n// ============================================================================\n\n/** Maximum file size to index (1 MB). */\nconst MAX_FILE_SIZE = 1024 * 1024;\n\n/** Directories unconditionally skipped during traversal. */\nconst SKIP_DIRS = new Set([\n\t\"node_modules\",\n\t\".git\",\n\t\".dreb/index\",\n\t\".hg\",\n\t\".svn\",\n\t\"__pycache__\",\n\t\".tox\",\n\t\".venv\",\n\t\"dist\",\n\t\"build\",\n\t\".next\",\n\t\".nuxt\",\n\t\"coverage\",\n\t\".cache\",\n]);\n\n/** Extension → FileType mapping. */\nconst EXTENSION_MAP: ReadonlyMap<string, FileType> = new Map<string, FileType>([\n\t// Tree-sitter languages\n\t[\".ts\", \"typescript\"],\n\t[\".tsx\", \"tsx\"],\n\t[\".js\", \"javascript\"],\n\t[\".mjs\", \"javascript\"],\n\t[\".cjs\", \"javascript\"],\n\t[\".py\", \"python\"],\n\t[\".go\", \"go\"],\n\t[\".rs\", \"rust\"],\n\t[\".java\", \"java\"],\n\t[\".c\", \"c\"],\n\t[\".h\", \"c\"],\n\t[\".cpp\", \"cpp\"],\n\t[\".hpp\", \"cpp\"],\n\t[\".cc\", \"cpp\"],\n\t[\".cxx\", \"cpp\"],\n\t[\".hh\", \"cpp\"],\n\t[\".hxx\", \"cpp\"],\n\t// Text file types\n\t[\".md\", \"markdown\"],\n\t[\".mdx\", \"markdown\"],\n\t[\".yml\", \"yaml\"],\n\t[\".yaml\", \"yaml\"],\n\t[\".json\", \"json\"],\n\t[\".toml\", \"toml\"],\n\t[\".txt\", \"plaintext\"],\n\t[\".cfg\", \"plaintext\"],\n\t[\".ini\", \"plaintext\"],\n\t[\".env\", \"plaintext\"],\n\t[\".conf\", \"plaintext\"],\n]);\n\n// ============================================================================\n// Public API\n// ============================================================================\n\n/**\n * Detect the {@link FileType} for a file path based on its extension.\n * Returns `null` for unrecognized extensions or files without an extension.\n */\nexport function detectFileType(filePath: string): FileType | null {\n\tconst ext = extname(filePath).toLowerCase();\n\tif (!ext) return null;\n\treturn EXTENSION_MAP.get(ext) ?? null;\n}\n\n/**\n * Scan a project directory and return all indexable files.\n *\n * Walks the tree rooted at {@link projectRoot}, respects `.gitignore` rules,\n * skips binary / oversized files, and optionally includes memory files from\n * a global memory directory.\n */\nexport async function scanProject(projectRoot: string, globalMemoryDir?: string): Promise<ScannedFile[]> {\n\tconst results: ScannedFile[] = [];\n\n\t// Detect if projectRoot is the home directory — use shallow scan mode\n\t// to avoid recursing into the entire home dir (which would be catastrophic).\n\tconst isHomeDir = isHomeDirPath(projectRoot);\n\n\tif (isHomeDir) {\n\t\t// Shallow mode: only scan top-level files and ~/.dreb/memory/\n\t\tscanShallow(projectRoot, results);\n\t} else {\n\t\t// Normal mode: full recursive walk with .gitignore\n\t\tconst ig = ignore();\n\t\tloadGitignore(ig, projectRoot, projectRoot);\n\t\twalkDirectory(projectRoot, projectRoot, ig, results);\n\t}\n\n\t// Include global memory files if the directory exists\n\tif (globalMemoryDir && existsSync(globalMemoryDir)) {\n\t\tscanMemoryDir(globalMemoryDir, projectRoot, results);\n\t}\n\n\treturn results;\n}\n\n/** Check if a path is the user's home directory. */\nfunction isHomeDirPath(dir: string): boolean {\n\ttry {\n\t\tconst home = homedir();\n\t\t// Normalize trailing slashes for comparison\n\t\tconst normalizedDir = dir.replace(/[/\\\\]+$/, \"\");\n\t\tconst normalizedHome = home.replace(/[/\\\\]+$/, \"\");\n\t\treturn normalizedDir === normalizedHome;\n\t} catch {\n\t\treturn false;\n\t}\n}\n\n/**\n * Shallow scan mode for home directory: only index top-level files\n * (no directory recursion) to avoid scanning the entire home directory.\n * Memory files are handled separately via scanMemoryDir.\n */\nfunction scanShallow(dir: string, results: ScannedFile[]): void {\n\tlet entries: string[];\n\ttry {\n\t\tentries = readdirSync(dir);\n\t} catch {\n\t\treturn;\n\t}\n\n\tfor (const entry of entries) {\n\t\t// Skip dotfiles/dotdirs in home dir (except specific ones we want)\n\t\tif (entry.startsWith(\".\")) continue;\n\n\t\tconst fullPath = join(dir, entry);\n\n\t\tlet stats: Stats;\n\t\ttry {\n\t\t\tstats = statSync(fullPath);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\n\t\t// Only index files, not directories (shallow mode)\n\t\tif (!stats.isFile()) continue;\n\t\tif (stats.size > MAX_FILE_SIZE) continue;\n\t\tif (stats.size === 0) continue;\n\n\t\tconst fileType = detectFileType(entry);\n\t\tif (!fileType) continue;\n\n\t\tresults.push({\n\t\t\tfilePath: entry,\n\t\t\tfileType,\n\t\t\tmtime: stats.mtimeMs,\n\t\t});\n\t}\n}\n\n// ============================================================================\n// Internal helpers\n// ============================================================================\n\ntype IgnoreMatcher = ReturnType<typeof ignore>;\n\n/** Convert an OS path to posix separators for ignore matching. */\nfunction toPosix(p: string): string {\n\treturn p.split(sep).join(\"/\");\n}\n\n/** Load .gitignore rules from a directory into the ignore matcher. */\nfunction loadGitignore(ig: IgnoreMatcher, dir: string, root: string): void {\n\tconst gitignorePath = join(dir, \".gitignore\");\n\tif (!existsSync(gitignorePath)) return;\n\n\ttry {\n\t\tconst content = readFileSync(gitignorePath, \"utf-8\");\n\t\tconst relDir = relative(root, dir);\n\t\tconst prefix = relDir ? `${toPosix(relDir)}/` : \"\";\n\n\t\tconst patterns = content\n\t\t\t.split(/\\r?\\n/)\n\t\t\t.map((line) => prefixPattern(line, prefix))\n\t\t\t.filter((line): line is string => line !== null);\n\n\t\tif (patterns.length > 0) {\n\t\t\tig.add(patterns);\n\t\t}\n\t} catch {\n\t\t// Unreadable .gitignore — skip silently\n\t}\n}\n\n/**\n * Prefix a .gitignore pattern with a directory path so it applies\n * correctly when matching against root-relative paths.\n */\nfunction prefixPattern(line: string, prefix: string): string | null {\n\tconst trimmed = line.trim();\n\tif (!trimmed) return null;\n\tif (trimmed.startsWith(\"#\") && !trimmed.startsWith(\"\\\\#\")) return null;\n\n\tlet pattern = line;\n\tlet negated = false;\n\n\tif (pattern.startsWith(\"!\")) {\n\t\tnegated = true;\n\t\tpattern = pattern.slice(1);\n\t} else if (pattern.startsWith(\"\\\\!\")) {\n\t\tpattern = pattern.slice(1);\n\t}\n\n\tconst prefixed = prefix ? `${prefix}${pattern}` : pattern;\n\treturn negated ? `!${prefixed}` : prefixed;\n}\n\n/**\n * Check if a directory component (relative to root) should be unconditionally skipped.\n * Handles both top-level names (\"node_modules\") and nested paths (\".dreb/index\").\n */\nfunction shouldSkipDir(relPath: string): boolean {\n\tconst posix = toPosix(relPath);\n\n\t// Check the directory name itself\n\tconst parts = posix.split(\"/\");\n\tconst name = parts[parts.length - 1];\n\tif (SKIP_DIRS.has(name)) return true;\n\n\t// Check multi-segment skip patterns (e.g. \".dreb/index\")\n\tfor (const skip of SKIP_DIRS) {\n\t\tif (skip.includes(\"/\") && (posix === skip || posix.endsWith(`/${skip}`))) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n\n/** Recursively walk a directory, collecting indexable files. */\nfunction walkDirectory(dir: string, root: string, ig: IgnoreMatcher, results: ScannedFile[]): void {\n\tlet entries: string[];\n\ttry {\n\t\tentries = readdirSync(dir);\n\t} catch {\n\t\treturn; // Permission denied, etc.\n\t}\n\n\tfor (const entry of entries) {\n\t\tconst fullPath = join(dir, entry);\n\t\tconst relPath = relative(root, fullPath);\n\t\tconst posixRel = toPosix(relPath);\n\n\t\tlet stats: Stats;\n\t\ttry {\n\t\t\tstats = statSync(fullPath);\n\t\t} catch {\n\t\t\tcontinue; // Broken symlink, etc.\n\t\t}\n\n\t\tif (stats.isDirectory()) {\n\t\t\t// Hard-coded skip list\n\t\t\tif (shouldSkipDir(relPath)) continue;\n\n\t\t\t// .gitignore check (directories need trailing slash)\n\t\t\tif (ig.ignores(`${posixRel}/`)) continue;\n\n\t\t\t// Load nested .gitignore before descending\n\t\t\tloadGitignore(ig, fullPath, root);\n\n\t\t\twalkDirectory(fullPath, root, ig, results);\n\t\t\tcontinue;\n\t\t}\n\n\t\tif (!stats.isFile()) continue;\n\n\t\t// .gitignore check for files\n\t\tif (ig.ignores(posixRel)) continue;\n\n\t\t// Size gate\n\t\tif (stats.size > MAX_FILE_SIZE) continue;\n\t\tif (stats.size === 0) continue;\n\n\t\t// File type detection\n\t\tconst fileType = detectFileType(entry);\n\t\tif (!fileType) continue;\n\n\t\tresults.push({\n\t\t\tfilePath: posixRel,\n\t\t\tfileType,\n\t\t\tmtime: stats.mtimeMs,\n\t\t});\n\t}\n}\n\n/**\n * Scan a memory directory (project or global) for indexable files.\n *\n * Memory directories are always fully included — no .gitignore filtering —\n * because they live outside the normal project tree or in `.dreb/` which\n * is typically gitignored.\n *\n * Paths for global memory files are stored with a `~memory/` prefix\n * to distinguish them from project files.\n */\nfunction scanMemoryDir(memoryDir: string, projectRoot: string, results: ScannedFile[], baseMemoryDir?: string): void {\n\tlet entries: string[];\n\ttry {\n\t\tentries = readdirSync(memoryDir);\n\t} catch {\n\t\treturn;\n\t}\n\n\tfor (const entry of entries) {\n\t\tconst fullPath = join(memoryDir, entry);\n\n\t\tlet stats: Stats;\n\t\ttry {\n\t\t\tstats = statSync(fullPath);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\n\t\tif (stats.isDirectory()) {\n\t\t\t// Recurse into subdirectories\n\t\t\tscanMemoryDir(fullPath, projectRoot, results, baseMemoryDir ?? memoryDir);\n\t\t\tcontinue;\n\t\t}\n\n\t\tif (!stats.isFile()) continue;\n\t\tif (stats.size > MAX_FILE_SIZE) continue;\n\t\tif (stats.size === 0) continue;\n\n\t\tconst fileType = detectFileType(entry);\n\t\tif (!fileType) continue;\n\n\t\t// If the memory dir is inside the project root, use normal relative path.\n\t\t// Otherwise, use a ~memory/ prefix so paths remain unique and identifiable.\n\t\tconst rel = relative(projectRoot, fullPath);\n\t\tconst isOutsideProject = rel.startsWith(\"..\") || isAbsolute(rel);\n\t\tconst rootMemoryDir = baseMemoryDir ?? memoryDir;\n\t\tconst filePath = isOutsideProject ? `~memory/${relative(rootMemoryDir, fullPath)}` : rel;\n\n\t\tresults.push({\n\t\t\tfilePath: toPosix(filePath),\n\t\t\tfileType,\n\t\t\tmtime: stats.mtimeMs,\n\t\t});\n\t}\n}\n"]}

package/dist/core/search/scanner.js

@@ -1,335 +0,0 @@
-/**
- * File scanner for the semantic search subsystem.
- *
- * Discovers project files for indexing by walking the directory tree,
- * respecting .gitignore rules, and classifying files by type.
- */
-import { existsSync, readdirSync, readFileSync, statSync } from "node:fs";
-import { homedir } from "node:os";
-import { extname, isAbsolute, join, relative, sep } from "node:path";
-import ignore from "ignore";
-// ============================================================================
-// Constants
-// ============================================================================
-/** Maximum file size to index (1 MB). */
-const MAX_FILE_SIZE = 1024 * 1024;
-/** Directories unconditionally skipped during traversal. */
-const SKIP_DIRS = new Set([
-    "node_modules",
-    ".git",
-    ".dreb/index",
-    ".hg",
-    ".svn",
-    "__pycache__",
-    ".tox",
-    ".venv",
-    "dist",
-    "build",
-    ".next",
-    ".nuxt",
-    "coverage",
-    ".cache",
-]);
-/** Extension → FileType mapping. */
-const EXTENSION_MAP = new Map([
-    // Tree-sitter languages
-    [".ts", "typescript"],
-    [".tsx", "tsx"],
-    [".js", "javascript"],
-    [".mjs", "javascript"],
-    [".cjs", "javascript"],
-    [".py", "python"],
-    [".go", "go"],
-    [".rs", "rust"],
-    [".java", "java"],
-    [".c", "c"],
-    [".h", "c"],
-    [".cpp", "cpp"],
-    [".hpp", "cpp"],
-    [".cc", "cpp"],
-    [".cxx", "cpp"],
-    [".hh", "cpp"],
-    [".hxx", "cpp"],
-    // Text file types
-    [".md", "markdown"],
-    [".mdx", "markdown"],
-    [".yml", "yaml"],
-    [".yaml", "yaml"],
-    [".json", "json"],
-    [".toml", "toml"],
-    [".txt", "plaintext"],
-    [".cfg", "plaintext"],
-    [".ini", "plaintext"],
-    [".env", "plaintext"],
-    [".conf", "plaintext"],
-]);
-// ============================================================================
-// Public API
-// ============================================================================
-/**
- * Detect the {@link FileType} for a file path based on its extension.
- * Returns `null` for unrecognized extensions or files without an extension.
- */
-export function detectFileType(filePath) {
-    const ext = extname(filePath).toLowerCase();
-    if (!ext)
-        return null;
-    return EXTENSION_MAP.get(ext) ?? null;
-}
-/**
- * Scan a project directory and return all indexable files.
- *
- * Walks the tree rooted at {@link projectRoot}, respects `.gitignore` rules,
- * skips binary / oversized files, and optionally includes memory files from
- * a global memory directory.
- */
-export async function scanProject(projectRoot, globalMemoryDir) {
-    const results = [];
-    // Detect if projectRoot is the home directory — use shallow scan mode
-    // to avoid recursing into the entire home dir (which would be catastrophic).
-    const isHomeDir = isHomeDirPath(projectRoot);
-    if (isHomeDir) {
-        // Shallow mode: only scan top-level files and ~/.dreb/memory/
-        scanShallow(projectRoot, results);
-    }
-    else {
-        // Normal mode: full recursive walk with .gitignore
-        const ig = ignore();
-        loadGitignore(ig, projectRoot, projectRoot);
-        walkDirectory(projectRoot, projectRoot, ig, results);
-    }
-    // Include global memory files if the directory exists
-    if (globalMemoryDir && existsSync(globalMemoryDir)) {
-        scanMemoryDir(globalMemoryDir, projectRoot, results);
-    }
-    return results;
-}
-/** Check if a path is the user's home directory. */
-function isHomeDirPath(dir) {
-    try {
-        const home = homedir();
-        // Normalize trailing slashes for comparison
-        const normalizedDir = dir.replace(/[/\\]+$/, "");
-        const normalizedHome = home.replace(/[/\\]+$/, "");
-        return normalizedDir === normalizedHome;
-    }
-    catch {
-        return false;
-    }
-}
-/**
- * Shallow scan mode for home directory: only index top-level files
- * (no directory recursion) to avoid scanning the entire home directory.
- * Memory files are handled separately via scanMemoryDir.
- */
-function scanShallow(dir, results) {
-    let entries;
-    try {
-        entries = readdirSync(dir);
-    }
-    catch {
-        return;
-    }
-    for (const entry of entries) {
-        // Skip dotfiles/dotdirs in home dir (except specific ones we want)
-        if (entry.startsWith("."))
-            continue;
-        const fullPath = join(dir, entry);
-        let stats;
-        try {
-            stats = statSync(fullPath);
-        }
-        catch {
-            continue;
-        }
-        // Only index files, not directories (shallow mode)
-        if (!stats.isFile())
-            continue;
-        if (stats.size > MAX_FILE_SIZE)
-            continue;
-        if (stats.size === 0)
-            continue;
-        const fileType = detectFileType(entry);
-        if (!fileType)
-            continue;
-        results.push({
-            filePath: entry,
-            fileType,
-            mtime: stats.mtimeMs,
-        });
-    }
-}
-/** Convert an OS path to posix separators for ignore matching. */
-function toPosix(p) {
-    return p.split(sep).join("/");
-}
-/** Load .gitignore rules from a directory into the ignore matcher. */
-function loadGitignore(ig, dir, root) {
-    const gitignorePath = join(dir, ".gitignore");
-    if (!existsSync(gitignorePath))
-        return;
-    try {
-        const content = readFileSync(gitignorePath, "utf-8");
-        const relDir = relative(root, dir);
-        const prefix = relDir ? `${toPosix(relDir)}/` : "";
-        const patterns = content
-            .split(/\r?\n/)
-            .map((line) => prefixPattern(line, prefix))
-            .filter((line) => line !== null);
-        if (patterns.length > 0) {
-            ig.add(patterns);
-        }
-    }
-    catch {
-        // Unreadable .gitignore — skip silently
-    }
-}
-/**
- * Prefix a .gitignore pattern with a directory path so it applies
- * correctly when matching against root-relative paths.
- */
-function prefixPattern(line, prefix) {
-    const trimmed = line.trim();
-    if (!trimmed)
-        return null;
-    if (trimmed.startsWith("#") && !trimmed.startsWith("\\#"))
-        return null;
-    let pattern = line;
-    let negated = false;
-    if (pattern.startsWith("!")) {
-        negated = true;
-        pattern = pattern.slice(1);
-    }
-    else if (pattern.startsWith("\\!")) {
-        pattern = pattern.slice(1);
-    }
-    const prefixed = prefix ? `${prefix}${pattern}` : pattern;
-    return negated ? `!${prefixed}` : prefixed;
-}
-/**
- * Check if a directory component (relative to root) should be unconditionally skipped.
- * Handles both top-level names ("node_modules") and nested paths (".dreb/index").
- */
-function shouldSkipDir(relPath) {
-    const posix = toPosix(relPath);
-    // Check the directory name itself
-    const parts = posix.split("/");
-    const name = parts[parts.length - 1];
-    if (SKIP_DIRS.has(name))
-        return true;
-    // Check multi-segment skip patterns (e.g. ".dreb/index")
-    for (const skip of SKIP_DIRS) {
-        if (skip.includes("/") && (posix === skip || posix.endsWith(`/${skip}`))) {
-            return true;
-        }
-    }
-    return false;
-}
-/** Recursively walk a directory, collecting indexable files. */
-function walkDirectory(dir, root, ig, results) {
-    let entries;
-    try {
-        entries = readdirSync(dir);
-    }
-    catch {
-        return; // Permission denied, etc.
-    }
-    for (const entry of entries) {
-        const fullPath = join(dir, entry);
-        const relPath = relative(root, fullPath);
-        const posixRel = toPosix(relPath);
-        let stats;
-        try {
-            stats = statSync(fullPath);
-        }
-        catch {
-            continue; // Broken symlink, etc.
-        }
-        if (stats.isDirectory()) {
-            // Hard-coded skip list
-            if (shouldSkipDir(relPath))
-                continue;
-            // .gitignore check (directories need trailing slash)
-            if (ig.ignores(`${posixRel}/`))
-                continue;
-            // Load nested .gitignore before descending
-            loadGitignore(ig, fullPath, root);
-            walkDirectory(fullPath, root, ig, results);
-            continue;
-        }
-        if (!stats.isFile())
-            continue;
-        // .gitignore check for files
-        if (ig.ignores(posixRel))
-            continue;
-        // Size gate
-        if (stats.size > MAX_FILE_SIZE)
-            continue;
-        if (stats.size === 0)
-            continue;
-        // File type detection
-        const fileType = detectFileType(entry);
-        if (!fileType)
-            continue;
-        results.push({
-            filePath: posixRel,
-            fileType,
-            mtime: stats.mtimeMs,
-        });
-    }
-}
-/**
- * Scan a memory directory (project or global) for indexable files.
- *
- * Memory directories are always fully included — no .gitignore filtering —
- * because they live outside the normal project tree or in `.dreb/` which
- * is typically gitignored.
- *
- * Paths for global memory files are stored with a `~memory/` prefix
- * to distinguish them from project files.
- */
-function scanMemoryDir(memoryDir, projectRoot, results, baseMemoryDir) {
-    let entries;
-    try {
-        entries = readdirSync(memoryDir);
-    }
-    catch {
-        return;
-    }
-    for (const entry of entries) {
-        const fullPath = join(memoryDir, entry);
-        let stats;
-        try {
-            stats = statSync(fullPath);
-        }
-        catch {
-            continue;
-        }
-        if (stats.isDirectory()) {
-            // Recurse into subdirectories
-            scanMemoryDir(fullPath, projectRoot, results, baseMemoryDir ?? memoryDir);
-            continue;
-        }
-        if (!stats.isFile())
-            continue;
-        if (stats.size > MAX_FILE_SIZE)
-            continue;
-        if (stats.size === 0)
-            continue;
-        const fileType = detectFileType(entry);
-        if (!fileType)
-            continue;
-        // If the memory dir is inside the project root, use normal relative path.
-        // Otherwise, use a ~memory/ prefix so paths remain unique and identifiable.
-        const rel = relative(projectRoot, fullPath);
-        const isOutsideProject = rel.startsWith("..") || isAbsolute(rel);
-        const rootMemoryDir = baseMemoryDir ?? memoryDir;
-        const filePath = isOutsideProject ? `~memory/${relative(rootMemoryDir, fullPath)}` : rel;
-        results.push({
-            filePath: toPosix(filePath),
-            fileType,
-            mtime: stats.mtimeMs,
-        });
-    }
-}
-//# sourceMappingURL=scanner.js.map

package/dist/core/search/scanner.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"scanner.js","sourceRoot":"","sources":["../../../src/core/search/scanner.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,EAAE,UAAU,EAAE,WAAW,EAAE,YAAY,EAAc,QAAQ,EAAE,MAAM,SAAS,CAAC;AACtF,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,MAAM,WAAW,CAAC;AACrE,OAAO,MAAM,MAAM,QAAQ,CAAC;AAiB5B,+EAA+E;AAC/E,YAAY;AACZ,+EAA+E;AAE/E,yCAAyC;AACzC,MAAM,aAAa,GAAG,IAAI,GAAG,IAAI,CAAC;AAElC,4DAA4D;AAC5D,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC;IACzB,cAAc;IACd,MAAM;IACN,aAAa;IACb,KAAK;IACL,MAAM;IACN,aAAa;IACb,MAAM;IACN,OAAO;IACP,MAAM;IACN,OAAO;IACP,OAAO;IACP,OAAO;IACP,UAAU;IACV,QAAQ;CACR,CAAC,CAAC;AAEH,sCAAoC;AACpC,MAAM,aAAa,GAAkC,IAAI,GAAG,CAAmB;IAC9E,wBAAwB;IACxB,CAAC,KAAK,EAAE,YAAY,CAAC;IACrB,CAAC,MAAM,EAAE,KAAK,CAAC;IACf,CAAC,KAAK,EAAE,YAAY,CAAC;IACrB,CAAC,MAAM,EAAE,YAAY,CAAC;IACtB,CAAC,MAAM,EAAE,YAAY,CAAC;IACtB,CAAC,KAAK,EAAE,QAAQ,CAAC;IACjB,CAAC,KAAK,EAAE,IAAI,CAAC;IACb,CAAC,KAAK,EAAE,MAAM,CAAC;IACf,CAAC,OAAO,EAAE,MAAM,CAAC;IACjB,CAAC,IAAI,EAAE,GAAG,CAAC;IACX,CAAC,IAAI,EAAE,GAAG,CAAC;IACX,CAAC,MAAM,EAAE,KAAK,CAAC;IACf,CAAC,MAAM,EAAE,KAAK,CAAC;IACf,CAAC,KAAK,EAAE,KAAK,CAAC;IACd,CAAC,MAAM,EAAE,KAAK,CAAC;IACf,CAAC,KAAK,EAAE,KAAK,CAAC;IACd,CAAC,MAAM,EAAE,KAAK,CAAC;IACf,kBAAkB;IAClB,CAAC,KAAK,EAAE,UAAU,CAAC;IACnB,CAAC,MAAM,EAAE,UAAU,CAAC;IACpB,CAAC,MAAM,EAAE,MAAM,CAAC;IAChB,CAAC,OAAO,EAAE,MAAM,CAAC;IACjB,CAAC,OAAO,EAAE,MAAM,CAAC;IACjB,CAAC,OAAO,EAAE,MAAM,CAAC;IACjB,CAAC,MAAM,EAAE,WAAW,CAAC;IACrB,CAAC,MAAM,EAAE,WAAW,CAAC;IACrB,CAAC,MAAM,EAAE,WAAW,CAAC;IACrB,CAAC,MAAM,EAAE,WAAW,CAAC;IACrB,CAAC,OAAO,EAAE,WAAW,CAAC;CACtB,CAAC,CAAC;AAEH,+EAA+E;AAC/E,aAAa;AACb,+EAA+E;AAE/E;;;GAGG;AACH,MAAM,UAAU,cAAc,CAAC,QAAgB,EAAmB;IACjE,MAAM,GAAG,GAAG,OAAO,CAAC,QAAQ,CAAC,CAAC,WAAW,EAAE,CAAC;IAC5C,IAAI,CAAC,GAAG;QAAE,OAAO,IAAI,CAAC;IACtB,OAAO,aAAa,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC;AAAA,CACtC;AAED;;;;;;GAMG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAAC,WAAmB,EAAE,eAAwB,EAA0B;IACxG,MAAM,OAAO,GAAkB,EAAE,CAAC;IAElC,wEAAsE;IACtE,6EAA6E;IAC7E,MAAM,SAAS,GAAG,aAAa,CAAC,WAAW,CAAC,CAAC;IAE7C,IAAI,SAAS,EAAE,CAAC;QACf,8DAA8D;QAC9D,WAAW,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;IACnC,CAAC;SAAM,CAAC;QACP,mDAAmD;QACnD,MAAM,EAAE,GAAG,MAAM,EAAE,CAAC;QACpB,aAAa,CAAC,EAAE,EAAE,WAAW,EAAE,WAAW,CAAC,CAAC;QAC5C,aAAa,CAAC,WAAW,EAAE,WAAW,EAAE,EAAE,EAAE,OAAO,CAAC,CAAC;IACtD,CAAC;IAED,sDAAsD;IACtD,IAAI,eAAe,IAAI,UAAU,CAAC,eAAe,CAAC,EAAE,CAAC;QACpD,aAAa,CAAC,eAAe,EAAE,WAAW,EAAE,OAAO,CAAC,CAAC;IACtD,CAAC;IAED,OAAO,OAAO,CAAC;AAAA,CACf;AAED,oDAAoD;AACpD,SAAS,aAAa,CAAC,GAAW,EAAW;IAC5C,IAAI,CAAC;QACJ,MAAM,IAAI,GAAG,OAAO,EAAE,CAAC;QACvB,4CAA4C;QAC5C,MAAM,aAAa,GAAG,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC;QACjD,MAAM,cAAc,GAAG,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,EAAE,CAAC,CAAC;QACnD,OAAO,aAAa,KAAK,cAAc,CAAC;IACzC,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,KAAK,CAAC;IACd,CAAC;AAAA,CACD;AAED;;;;GAIG;AACH,SAAS,WAAW,CAAC,GAAW,EAAE,OAAsB,EAAQ;IAC/D,IAAI,OAAiB,CAAC;IACtB,IAAI,CAAC;QACJ,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;IAC5B,CAAC;IAAC,MAAM,CAAC;QACR,OAAO;IACR,CAAC;IAED,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,mEAAmE;QACnE,IAAI,KAAK,CAAC,UAAU,CAAC,GAAG,CAAC;YAAE,SAAS;QAEpC,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QAElC,IAAI,KAAY,CAAC;QACjB,IAAI,CAAC;YACJ,KAAK,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC;QAC5B,CAAC;QAAC,MAAM,CAAC;YACR,SAAS;QACV,CAAC;QAED,mDAAmD;QACnD,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE;YAAE,SAAS;QAC9B,IAAI,KAAK,CAAC,IAAI,GAAG,aAAa;YAAE,SAAS;QACzC,IAAI,KAAK,CAAC,IAAI,KAAK,CAAC;YAAE,SAAS;QAE/B,MAAM,QAAQ,GAAG,cAAc,CAAC,KAAK,CAAC,CAAC;QACvC,IAAI,CAAC,QAAQ;YAAE,SAAS;QAExB,OAAO,CAAC,IAAI,CAAC;YACZ,QAAQ,EAAE,KAAK;YACf,QAAQ;YACR,KAAK,EAAE,KAAK,CAAC,OAAO;SACpB,CAAC,CAAC;IACJ,CAAC;AAAA,CACD;AAQD,kEAAkE;AAClE,SAAS,OAAO,CAAC,CAAS,EAAU;IACnC,OAAO,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;AAAA,CAC9B;AAED,sEAAsE;AACtE,SAAS,aAAa,CAAC,EAAiB,EAAE,GAAW,EAAE,IAAY,EAAQ;IAC1E,MAAM,aAAa,GAAG,IAAI,CAAC,GAAG,EAAE,YAAY,CAAC,CAAC;IAC9C,IAAI,CAAC,UAAU,CAAC,aAAa,CAAC;QAAE,OAAO;IAEvC,IAAI,CAAC;QACJ,MAAM,OAAO,GAAG,YAAY,CAAC,aAAa,EAAE,OAAO,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,QAAQ,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;QACnC,MAAM,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC;QAEnD,MAAM,QAAQ,GAAG,OAAO;aACtB,KAAK,CAAC,OAAO,CAAC;aACd,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,aAAa,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;aAC1C,MAAM,CAAC,CAAC,IAAI,EAAkB,EAAE,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC;QAElD,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACzB,EAAE,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;QAClB,CAAC;IACF,CAAC;IAAC,MAAM,CAAC;QACR,0CAAwC;IACzC,CAAC;AAAA,CACD;AAED;;;GAGG;AACH,SAAS,aAAa,CAAC,IAAY,EAAE,MAAc,EAAiB;IACnE,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,IAAI,CAAC,OAAO;QAAE,OAAO,IAAI,CAAC;IAC1B,IAAI,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IAEvE,IAAI,OAAO,GAAG,IAAI,CAAC;IACnB,IAAI,OAAO,GAAG,KAAK,CAAC;IAEpB,IAAI,OAAO,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;QAC7B,OAAO,GAAG,IAAI,CAAC;QACf,OAAO,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IAC5B,CAAC;SAAM,IAAI,OAAO,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;QACtC,OAAO,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IAC5B,CAAC;IAED,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAC,GAAG,MAAM,GAAG,OAAO,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC;IAC1D,OAAO,OAAO,CAAC,CAAC,CAAC,IAAI,QAAQ,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC;AAAA,CAC3C;AAED;;;GAGG;AACH,SAAS,aAAa,CAAC,OAAe,EAAW;IAChD,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IAE/B,kCAAkC;IAClC,MAAM,KAAK,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC/B,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IACrC,IAAI,SAAS,CAAC,GAAG,CAAC,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC;IAErC,yDAAyD;IACzD,KAAK,MAAM,IAAI,IAAI,SAAS,EAAE,CAAC;QAC9B,IAAI,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,KAAK,KAAK,IAAI,IAAI,KAAK,CAAC,QAAQ,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,EAAE,CAAC;YAC1E,OAAO,IAAI,CAAC;QACb,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC;AAAA,CACb;AAED,gEAAgE;AAChE,SAAS,aAAa,CAAC,GAAW,EAAE,IAAY,EAAE,EAAiB,EAAE,OAAsB,EAAQ;IAClG,IAAI,OAAiB,CAAC;IACtB,IAAI,CAAC;QACJ,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;IAC5B,CAAC;IAAC,MAAM,CAAC;QACR,OAAO,CAAC,0BAA0B;IACnC,CAAC;IAED,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QAClC,MAAM,OAAO,GAAG,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;QACzC,MAAM,QAAQ,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;QAElC,IAAI,KAAY,CAAC;QACjB,IAAI,CAAC;YACJ,KAAK,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC;QAC5B,CAAC;QAAC,MAAM,CAAC;YACR,SAAS,CAAC,uBAAuB;QAClC,CAAC;QAED,IAAI,KAAK,CAAC,WAAW,EAAE,EAAE,CAAC;YACzB,uBAAuB;YACvB,IAAI,aAAa,CAAC,OAAO,CAAC;gBAAE,SAAS;YAErC,qDAAqD;YACrD,IAAI,EAAE,CAAC,OAAO,CAAC,GAAG,QAAQ,GAAG,CAAC;gBAAE,SAAS;YAEzC,2CAA2C;YAC3C,aAAa,CAAC,EAAE,EAAE,QAAQ,EAAE,IAAI,CAAC,CAAC;YAElC,aAAa,CAAC,QAAQ,EAAE,IAAI,EAAE,EAAE,EAAE,OAAO,CAAC,CAAC;YAC3C,SAAS;QACV,CAAC;QAED,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE;YAAE,SAAS;QAE9B,6BAA6B;QAC7B,IAAI,EAAE,CAAC,OAAO,CAAC,QAAQ,CAAC;YAAE,SAAS;QAEnC,YAAY;QACZ,IAAI,KAAK,CAAC,IAAI,GAAG,aAAa;YAAE,SAAS;QACzC,IAAI,KAAK,CAAC,IAAI,KAAK,CAAC;YAAE,SAAS;QAE/B,sBAAsB;QACtB,MAAM,QAAQ,GAAG,cAAc,CAAC,KAAK,CAAC,CAAC;QACvC,IAAI,CAAC,QAAQ;YAAE,SAAS;QAExB,OAAO,CAAC,IAAI,CAAC;YACZ,QAAQ,EAAE,QAAQ;YAClB,QAAQ;YACR,KAAK,EAAE,KAAK,CAAC,OAAO;SACpB,CAAC,CAAC;IACJ,CAAC;AAAA,CACD;AAED;;;;;;;;;GASG;AACH,SAAS,aAAa,CAAC,SAAiB,EAAE,WAAmB,EAAE,OAAsB,EAAE,aAAsB,EAAQ;IACpH,IAAI,OAAiB,CAAC;IACtB,IAAI,CAAC;QACJ,OAAO,GAAG,WAAW,CAAC,SAAS,CAAC,CAAC;IAClC,CAAC;IAAC,MAAM,CAAC;QACR,OAAO;IACR,CAAC;IAED,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;QAExC,IAAI,KAAY,CAAC;QACjB,IAAI,CAAC;YACJ,KAAK,GAAG,QAAQ,CAAC,QAAQ,CAAC,CAAC;QAC5B,CAAC;QAAC,MAAM,CAAC;YACR,SAAS;QACV,CAAC;QAED,IAAI,KAAK,CAAC,WAAW,EAAE,EAAE,CAAC;YACzB,8BAA8B;YAC9B,aAAa,CAAC,QAAQ,EAAE,WAAW,EAAE,OAAO,EAAE,aAAa,IAAI,SAAS,CAAC,CAAC;YAC1E,SAAS;QACV,CAAC;QAED,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE;YAAE,SAAS;QAC9B,IAAI,KAAK,CAAC,IAAI,GAAG,aAAa;YAAE,SAAS;QACzC,IAAI,KAAK,CAAC,IAAI,KAAK,CAAC;YAAE,SAAS;QAE/B,MAAM,QAAQ,GAAG,cAAc,CAAC,KAAK,CAAC,CAAC;QACvC,IAAI,CAAC,QAAQ;YAAE,SAAS;QAExB,0EAA0E;QAC1E,4EAA4E;QAC5E,MAAM,GAAG,GAAG,QAAQ,CAAC,WAAW,EAAE,QAAQ,CAAC,CAAC;QAC5C,MAAM,gBAAgB,GAAG,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,UAAU,CAAC,GAAG,CAAC,CAAC;QACjE,MAAM,aAAa,GAAG,aAAa,IAAI,SAAS,CAAC;QACjD,MAAM,QAAQ,GAAG,gBAAgB,CAAC,CAAC,CAAC,WAAW,QAAQ,CAAC,aAAa,EAAE,QAAQ,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC;QAEzF,OAAO,CAAC,IAAI,CAAC;YACZ,QAAQ,EAAE,OAAO,CAAC,QAAQ,CAAC;YAC3B,QAAQ;YACR,KAAK,EAAE,KAAK,CAAC,OAAO;SACpB,CAAC,CAAC;IACJ,CAAC;AAAA,CACD","sourcesContent":["/**\n * File scanner for the semantic search subsystem.\n *\n * Discovers project files for indexing by walking the directory tree,\n * respecting .gitignore rules, and classifying files by type.\n */\n\nimport { existsSync, readdirSync, readFileSync, type Stats, statSync } from \"node:fs\";\nimport { homedir } from \"node:os\";\nimport { extname, isAbsolute, join, relative, sep } from \"node:path\";\nimport ignore from \"ignore\";\nimport type { FileType } from \"./types.js\";\n\n// ============================================================================\n// Public types\n// ============================================================================\n\n/** A file discovered by the scanner, ready for indexing. */\nexport interface ScannedFile {\n\t/** Path relative to the project root (posix separators). */\n\tfilePath: string;\n\t/** Detected file type. */\n\tfileType: FileType;\n\t/** File modification time in milliseconds since epoch. */\n\tmtime: number;\n}\n\n// ============================================================================\n// Constants\n// ============================================================================\n\n/** Maximum file size to index (1 MB). */\nconst MAX_FILE_SIZE = 1024 * 1024;\n\n/** Directories unconditionally skipped during traversal. */\nconst SKIP_DIRS = new Set([\n\t\"node_modules\",\n\t\".git\",\n\t\".dreb/index\",\n\t\".hg\",\n\t\".svn\",\n\t\"__pycache__\",\n\t\".tox\",\n\t\".venv\",\n\t\"dist\",\n\t\"build\",\n\t\".next\",\n\t\".nuxt\",\n\t\"coverage\",\n\t\".cache\",\n]);\n\n/** Extension → FileType mapping. */\nconst EXTENSION_MAP: ReadonlyMap<string, FileType> = new Map<string, FileType>([\n\t// Tree-sitter languages\n\t[\".ts\", \"typescript\"],\n\t[\".tsx\", \"tsx\"],\n\t[\".js\", \"javascript\"],\n\t[\".mjs\", \"javascript\"],\n\t[\".cjs\", \"javascript\"],\n\t[\".py\", \"python\"],\n\t[\".go\", \"go\"],\n\t[\".rs\", \"rust\"],\n\t[\".java\", \"java\"],\n\t[\".c\", \"c\"],\n\t[\".h\", \"c\"],\n\t[\".cpp\", \"cpp\"],\n\t[\".hpp\", \"cpp\"],\n\t[\".cc\", \"cpp\"],\n\t[\".cxx\", \"cpp\"],\n\t[\".hh\", \"cpp\"],\n\t[\".hxx\", \"cpp\"],\n\t// Text file types\n\t[\".md\", \"markdown\"],\n\t[\".mdx\", \"markdown\"],\n\t[\".yml\", \"yaml\"],\n\t[\".yaml\", \"yaml\"],\n\t[\".json\", \"json\"],\n\t[\".toml\", \"toml\"],\n\t[\".txt\", \"plaintext\"],\n\t[\".cfg\", \"plaintext\"],\n\t[\".ini\", \"plaintext\"],\n\t[\".env\", \"plaintext\"],\n\t[\".conf\", \"plaintext\"],\n]);\n\n// ============================================================================\n// Public API\n// ============================================================================\n\n/**\n * Detect the {@link FileType} for a file path based on its extension.\n * Returns `null` for unrecognized extensions or files without an extension.\n */\nexport function detectFileType(filePath: string): FileType | null {\n\tconst ext = extname(filePath).toLowerCase();\n\tif (!ext) return null;\n\treturn EXTENSION_MAP.get(ext) ?? null;\n}\n\n/**\n * Scan a project directory and return all indexable files.\n *\n * Walks the tree rooted at {@link projectRoot}, respects `.gitignore` rules,\n * skips binary / oversized files, and optionally includes memory files from\n * a global memory directory.\n */\nexport async function scanProject(projectRoot: string, globalMemoryDir?: string): Promise<ScannedFile[]> {\n\tconst results: ScannedFile[] = [];\n\n\t// Detect if projectRoot is the home directory — use shallow scan mode\n\t// to avoid recursing into the entire home dir (which would be catastrophic).\n\tconst isHomeDir = isHomeDirPath(projectRoot);\n\n\tif (isHomeDir) {\n\t\t// Shallow mode: only scan top-level files and ~/.dreb/memory/\n\t\tscanShallow(projectRoot, results);\n\t} else {\n\t\t// Normal mode: full recursive walk with .gitignore\n\t\tconst ig = ignore();\n\t\tloadGitignore(ig, projectRoot, projectRoot);\n\t\twalkDirectory(projectRoot, projectRoot, ig, results);\n\t}\n\n\t// Include global memory files if the directory exists\n\tif (globalMemoryDir && existsSync(globalMemoryDir)) {\n\t\tscanMemoryDir(globalMemoryDir, projectRoot, results);\n\t}\n\n\treturn results;\n}\n\n/** Check if a path is the user's home directory. */\nfunction isHomeDirPath(dir: string): boolean {\n\ttry {\n\t\tconst home = homedir();\n\t\t// Normalize trailing slashes for comparison\n\t\tconst normalizedDir = dir.replace(/[/\\\\]+$/, \"\");\n\t\tconst normalizedHome = home.replace(/[/\\\\]+$/, \"\");\n\t\treturn normalizedDir === normalizedHome;\n\t} catch {\n\t\treturn false;\n\t}\n}\n\n/**\n * Shallow scan mode for home directory: only index top-level files\n * (no directory recursion) to avoid scanning the entire home directory.\n * Memory files are handled separately via scanMemoryDir.\n */\nfunction scanShallow(dir: string, results: ScannedFile[]): void {\n\tlet entries: string[];\n\ttry {\n\t\tentries = readdirSync(dir);\n\t} catch {\n\t\treturn;\n\t}\n\n\tfor (const entry of entries) {\n\t\t// Skip dotfiles/dotdirs in home dir (except specific ones we want)\n\t\tif (entry.startsWith(\".\")) continue;\n\n\t\tconst fullPath = join(dir, entry);\n\n\t\tlet stats: Stats;\n\t\ttry {\n\t\t\tstats = statSync(fullPath);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\n\t\t// Only index files, not directories (shallow mode)\n\t\tif (!stats.isFile()) continue;\n\t\tif (stats.size > MAX_FILE_SIZE) continue;\n\t\tif (stats.size === 0) continue;\n\n\t\tconst fileType = detectFileType(entry);\n\t\tif (!fileType) continue;\n\n\t\tresults.push({\n\t\t\tfilePath: entry,\n\t\t\tfileType,\n\t\t\tmtime: stats.mtimeMs,\n\t\t});\n\t}\n}\n\n// ============================================================================\n// Internal helpers\n// ============================================================================\n\ntype IgnoreMatcher = ReturnType<typeof ignore>;\n\n/** Convert an OS path to posix separators for ignore matching. */\nfunction toPosix(p: string): string {\n\treturn p.split(sep).join(\"/\");\n}\n\n/** Load .gitignore rules from a directory into the ignore matcher. */\nfunction loadGitignore(ig: IgnoreMatcher, dir: string, root: string): void {\n\tconst gitignorePath = join(dir, \".gitignore\");\n\tif (!existsSync(gitignorePath)) return;\n\n\ttry {\n\t\tconst content = readFileSync(gitignorePath, \"utf-8\");\n\t\tconst relDir = relative(root, dir);\n\t\tconst prefix = relDir ? `${toPosix(relDir)}/` : \"\";\n\n\t\tconst patterns = content\n\t\t\t.split(/\\r?\\n/)\n\t\t\t.map((line) => prefixPattern(line, prefix))\n\t\t\t.filter((line): line is string => line !== null);\n\n\t\tif (patterns.length > 0) {\n\t\t\tig.add(patterns);\n\t\t}\n\t} catch {\n\t\t// Unreadable .gitignore — skip silently\n\t}\n}\n\n/**\n * Prefix a .gitignore pattern with a directory path so it applies\n * correctly when matching against root-relative paths.\n */\nfunction prefixPattern(line: string, prefix: string): string | null {\n\tconst trimmed = line.trim();\n\tif (!trimmed) return null;\n\tif (trimmed.startsWith(\"#\") && !trimmed.startsWith(\"\\\\#\")) return null;\n\n\tlet pattern = line;\n\tlet negated = false;\n\n\tif (pattern.startsWith(\"!\")) {\n\t\tnegated = true;\n\t\tpattern = pattern.slice(1);\n\t} else if (pattern.startsWith(\"\\\\!\")) {\n\t\tpattern = pattern.slice(1);\n\t}\n\n\tconst prefixed = prefix ? `${prefix}${pattern}` : pattern;\n\treturn negated ? `!${prefixed}` : prefixed;\n}\n\n/**\n * Check if a directory component (relative to root) should be unconditionally skipped.\n * Handles both top-level names (\"node_modules\") and nested paths (\".dreb/index\").\n */\nfunction shouldSkipDir(relPath: string): boolean {\n\tconst posix = toPosix(relPath);\n\n\t// Check the directory name itself\n\tconst parts = posix.split(\"/\");\n\tconst name = parts[parts.length - 1];\n\tif (SKIP_DIRS.has(name)) return true;\n\n\t// Check multi-segment skip patterns (e.g. \".dreb/index\")\n\tfor (const skip of SKIP_DIRS) {\n\t\tif (skip.includes(\"/\") && (posix === skip || posix.endsWith(`/${skip}`))) {\n\t\t\treturn true;\n\t\t}\n\t}\n\n\treturn false;\n}\n\n/** Recursively walk a directory, collecting indexable files. */\nfunction walkDirectory(dir: string, root: string, ig: IgnoreMatcher, results: ScannedFile[]): void {\n\tlet entries: string[];\n\ttry {\n\t\tentries = readdirSync(dir);\n\t} catch {\n\t\treturn; // Permission denied, etc.\n\t}\n\n\tfor (const entry of entries) {\n\t\tconst fullPath = join(dir, entry);\n\t\tconst relPath = relative(root, fullPath);\n\t\tconst posixRel = toPosix(relPath);\n\n\t\tlet stats: Stats;\n\t\ttry {\n\t\t\tstats = statSync(fullPath);\n\t\t} catch {\n\t\t\tcontinue; // Broken symlink, etc.\n\t\t}\n\n\t\tif (stats.isDirectory()) {\n\t\t\t// Hard-coded skip list\n\t\t\tif (shouldSkipDir(relPath)) continue;\n\n\t\t\t// .gitignore check (directories need trailing slash)\n\t\t\tif (ig.ignores(`${posixRel}/`)) continue;\n\n\t\t\t// Load nested .gitignore before descending\n\t\t\tloadGitignore(ig, fullPath, root);\n\n\t\t\twalkDirectory(fullPath, root, ig, results);\n\t\t\tcontinue;\n\t\t}\n\n\t\tif (!stats.isFile()) continue;\n\n\t\t// .gitignore check for files\n\t\tif (ig.ignores(posixRel)) continue;\n\n\t\t// Size gate\n\t\tif (stats.size > MAX_FILE_SIZE) continue;\n\t\tif (stats.size === 0) continue;\n\n\t\t// File type detection\n\t\tconst fileType = detectFileType(entry);\n\t\tif (!fileType) continue;\n\n\t\tresults.push({\n\t\t\tfilePath: posixRel,\n\t\t\tfileType,\n\t\t\tmtime: stats.mtimeMs,\n\t\t});\n\t}\n}\n\n/**\n * Scan a memory directory (project or global) for indexable files.\n *\n * Memory directories are always fully included — no .gitignore filtering —\n * because they live outside the normal project tree or in `.dreb/` which\n * is typically gitignored.\n *\n * Paths for global memory files are stored with a `~memory/` prefix\n * to distinguish them from project files.\n */\nfunction scanMemoryDir(memoryDir: string, projectRoot: string, results: ScannedFile[], baseMemoryDir?: string): void {\n\tlet entries: string[];\n\ttry {\n\t\tentries = readdirSync(memoryDir);\n\t} catch {\n\t\treturn;\n\t}\n\n\tfor (const entry of entries) {\n\t\tconst fullPath = join(memoryDir, entry);\n\n\t\tlet stats: Stats;\n\t\ttry {\n\t\t\tstats = statSync(fullPath);\n\t\t} catch {\n\t\t\tcontinue;\n\t\t}\n\n\t\tif (stats.isDirectory()) {\n\t\t\t// Recurse into subdirectories\n\t\t\tscanMemoryDir(fullPath, projectRoot, results, baseMemoryDir ?? memoryDir);\n\t\t\tcontinue;\n\t\t}\n\n\t\tif (!stats.isFile()) continue;\n\t\tif (stats.size > MAX_FILE_SIZE) continue;\n\t\tif (stats.size === 0) continue;\n\n\t\tconst fileType = detectFileType(entry);\n\t\tif (!fileType) continue;\n\n\t\t// If the memory dir is inside the project root, use normal relative path.\n\t\t// Otherwise, use a ~memory/ prefix so paths remain unique and identifiable.\n\t\tconst rel = relative(projectRoot, fullPath);\n\t\tconst isOutsideProject = rel.startsWith(\"..\") || isAbsolute(rel);\n\t\tconst rootMemoryDir = baseMemoryDir ?? memoryDir;\n\t\tconst filePath = isOutsideProject ? `~memory/${relative(rootMemoryDir, fullPath)}` : rel;\n\n\t\tresults.push({\n\t\t\tfilePath: toPosix(filePath),\n\t\t\tfileType,\n\t\t\tmtime: stats.mtimeMs,\n\t\t});\n\t}\n}\n"]}

package/dist/core/search/search.d.ts

@@ -1,42 +0,0 @@
-/**
- * Main search API.
- *
- * Orchestrates: check/build index → compute all 6 metrics → classify query
- * → duplicate columns → POEM rank → assemble results.
- */
-import type { IndexProgressCallback, SearchResult } from "./types.js";
-export interface SearchOptions {
-    /** Maximum number of results to return. Default: 20. */
-    limit?: number;
-    /** Restrict search to files under this path (relative to project root). */
-    pathFilter?: string;
-    /** Progress callback for indexing operations. */
-    onProgress?: IndexProgressCallback;
-}
-export declare class SearchEngine {
-    private readonly projectRoot;
-    private indexManager;
-    private embedder;
-    constructor(projectRoot: string);
-    /** Check if semantic search is available (requires node:sqlite). */
-    static isAvailable(): boolean;
-    /**
-     * Search the codebase with a natural language or identifier query.
-     *
-     * On first call, builds the index (scans, chunks, embeds). Subsequent calls
-     * incrementally update changed files before searching.
-     */
-    search(query: string, options?: SearchOptions): Promise<SearchResult[]>;
-    /** Get index stats without opening a new connection. */
-    getStats(): {
-        files: number;
-        chunks: number;
-    } | null;
-    /** Dispose resources. */
-    close(): void;
-    private getIndexManager;
-    private getIndexConfig;
-    private getOrCreateEmbedder;
-    private computeVectorScores;
-}
-//# sourceMappingURL=search.d.ts.map

package/dist/core/search/search.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"search.d.ts","sourceRoot":"","sources":["../../../src/core/search/search.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAcH,OAAO,KAAK,EAAe,qBAAqB,EAAgB,YAAY,EAAe,MAAM,YAAY,CAAC;AAe9G,MAAM,WAAW,aAAa;IAC7B,wDAAwD;IACxD,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,2EAA2E;IAC3E,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,iDAAiD;IACjD,UAAU,CAAC,EAAE,qBAAqB,CAAC;CACnC;AAMD,qBAAa,YAAY;IACxB,OAAO,CAAC,QAAQ,CAAC,WAAW,CAAS;IACrC,OAAO,CAAC,YAAY,CAA6B;IACjD,OAAO,CAAC,QAAQ,CAAyB;IAEzC,YAAY,WAAW,EAAE,MAAM,EAE9B;IAED,oEAAoE;IACpE,MAAM,CAAC,WAAW,IAAI,OAAO,CAE5B;IAED;;;;;OAKG;IACG,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,aAAa,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC,CAmH5E;IAED,wDAAwD;IACxD,QAAQ,IAAI;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,GAAG,IAAI,CAGnD;IAED,yBAAyB;IACzB,KAAK,IAAI,IAAI,CAKZ;IAMD,OAAO,CAAC,eAAe;IASvB,OAAO,CAAC,cAAc;YASR,mBAAmB;YAYnB,mBAAmB;CA4BjC","sourcesContent":["/**\n * Main search API.\n *\n * Orchestrates: check/build index → compute all 6 metrics → classify query\n * → duplicate columns → POEM rank → assemble results.\n */\n\nimport { homedir } from \"node:os\";\nimport path from \"node:path\";\nimport type { SearchDatabase } from \"./db.js\";\nimport { Embedder } from \"./embedder.js\";\nimport { IndexManager } from \"./index-manager.js\";\nimport { computeBm25Scores } from \"./metrics/bm25.js\";\nimport { computeGitRecencyScores } from \"./metrics/git-recency.js\";\nimport { computeImportGraphScores } from \"./metrics/import-graph.js\";\nimport { computePathMatchScores } from \"./metrics/path-match.js\";\nimport { computeSymbolMatchScores } from \"./metrics/symbol-match.js\";\nimport { poemRank } from \"./poem.js\";\nimport { classifyQuery } from \"./query-classifier.js\";\nimport type { IndexConfig, IndexProgressCallback, MetricScores, SearchResult, StoredChunk } from \"./types.js\";\nimport { topKSimilar } from \"./vector-store.js\";\n\n// ============================================================================\n// Constants\n// ============================================================================\n\nconst DEFAULT_MODEL_NAME = \"Xenova/all-MiniLM-L6-v2\";\nconst DEFAULT_RESULT_LIMIT = 20;\nconst METRIC_CANDIDATE_LIMIT = 1000;\n\n// ============================================================================\n// Search Options\n// ============================================================================\n\nexport interface SearchOptions {\n\t/** Maximum number of results to return. Default: 20. */\n\tlimit?: number;\n\t/** Restrict search to files under this path (relative to project root). */\n\tpathFilter?: string;\n\t/** Progress callback for indexing operations. */\n\tonProgress?: IndexProgressCallback;\n}\n\n// ============================================================================\n// Search Engine\n// ============================================================================\n\nexport class SearchEngine {\n\tprivate readonly projectRoot: string;\n\tprivate indexManager: IndexManager | null = null;\n\tprivate embedder: Embedder | null = null;\n\n\tconstructor(projectRoot: string) {\n\t\tthis.projectRoot = projectRoot;\n\t}\n\n\t/** Check if semantic search is available (requires node:sqlite). */\n\tstatic isAvailable(): boolean {\n\t\treturn IndexManager.isAvailable();\n\t}\n\n\t/**\n\t * Search the codebase with a natural language or identifier query.\n\t *\n\t * On first call, builds the index (scans, chunks, embeds). Subsequent calls\n\t * incrementally update changed files before searching.\n\t */\n\tasync search(query: string, options?: SearchOptions): Promise<SearchResult[]> {\n\t\tconst limit = options?.limit ?? DEFAULT_RESULT_LIMIT;\n\t\tconst onProgress = options?.onProgress;\n\n\t\t// Ensure index is built and up to date\n\t\tconst indexManager = this.getIndexManager();\n\t\tconst db = indexManager.getDb();\n\n\t\t// Share our embedder with IndexManager so it doesn't create a second one\n\t\tconst embedder = await this.getOrCreateEmbedder();\n\t\tindexManager.setEmbedder(embedder);\n\n\t\tawait indexManager.buildIndex(onProgress);\n\t\tawait indexManager.ensureEmbeddings(onProgress);\n\n\t\t// Get all chunks (potentially filtered by path)\n\t\tlet allChunks = db.getAllChunks();\n\t\tif (options?.pathFilter) {\n\t\t\tconst filter = options.pathFilter;\n\t\t\tallChunks = allChunks.filter((c) => c.filePath.startsWith(filter));\n\t\t}\n\n\t\tif (allChunks.length === 0) {\n\t\t\treturn [];\n\t\t}\n\n\t\t// Classify query type for POEM column weighting\n\t\tconst queryType = classifyQuery(query);\n\n\t\t// Compute all 6 metrics\n\t\tonProgress?.(\"searching\", 0, 6);\n\n\t\t// 1. BM25 (FTS5)\n\t\tconst bm25Scores = computeBm25Scores(db, sanitizeFtsQuery(query), METRIC_CANDIDATE_LIMIT);\n\t\tonProgress?.(\"searching\", 1, 6);\n\n\t\t// 2. Cosine similarity (vector search)\n\t\tconst cosineScores = await this.computeVectorScores(db, query, METRIC_CANDIDATE_LIMIT, onProgress);\n\t\tonProgress?.(\"searching\", 2, 6);\n\n\t\t// 3. Path match\n\t\tconst pathScores = computePathMatchScores(query, allChunks);\n\t\tonProgress?.(\"searching\", 3, 6);\n\n\t\t// 4. Symbol match\n\t\tconst symbols = db.getAllSymbols();\n\t\tconst symbolScores = computeSymbolMatchScores(query, symbols);\n\t\tonProgress?.(\"searching\", 4, 6);\n\n\t\t// 5. Import graph (use BM25 + cosine as seed scores, aggregated per file)\n\t\t// Only use files with strong scores as seeds — low-scoring files (e.g. from\n\t\t// common OR terms matching everywhere) pollute the seed set and prevent\n\t\t// meaningful propagation.\n\t\tconst fileSeedScores = aggregateFileScores(allChunks, bm25Scores, cosineScores);\n\t\tconst seedThreshold = computeSeedThreshold(fileSeedScores);\n\t\tconst filteredSeeds = new Map<number, number>();\n\t\tfor (const [fileId, score] of fileSeedScores) {\n\t\t\tif (score >= seedThreshold) filteredSeeds.set(fileId, score);\n\t\t}\n\t\tconst fileIdToChunkIds = buildFileChunkMap(allChunks);\n\t\tconst importScores = computeImportGraphScores(db, filteredSeeds, fileIdToChunkIds);\n\t\tonProgress?.(\"searching\", 5, 6);\n\n\t\t// 6. Git recency\n\t\tconst recencyScores = await computeGitRecencyScores(this.projectRoot, allChunks);\n\t\tonProgress?.(\"searching\", 6, 6);\n\n\t\t// Build MetricScores for each candidate chunk\n\t\tconst candidateIds = collectCandidateIds(\n\t\t\tbm25Scores,\n\t\t\tcosineScores,\n\t\t\tpathScores,\n\t\t\tsymbolScores,\n\t\t\timportScores,\n\t\t\trecencyScores,\n\t\t);\n\t\tconst candidates = new Map<number, MetricScores>();\n\n\t\tfor (const id of candidateIds) {\n\t\t\tcandidates.set(id, {\n\t\t\t\tbm25: bm25Scores.get(id) ?? 0,\n\t\t\t\tcosine: cosineScores.get(id) ?? 0,\n\t\t\t\tpathMatch: pathScores.get(id) ?? 0,\n\t\t\t\tsymbolMatch: symbolScores.get(id) ?? 0,\n\t\t\t\timportGraph: importScores.get(id) ?? 0,\n\t\t\t\tgitRecency: recencyScores.get(id) ?? 0,\n\t\t\t});\n\t\t}\n\n\t\tif (candidates.size === 0) {\n\t\t\treturn [];\n\t\t}\n\n\t\t// POEM rank\n\t\tconst ranked = poemRank(candidates, queryType);\n\n\t\t// Assemble results\n\t\tconst chunkMap = new Map<number, StoredChunk>();\n\t\tfor (const chunk of allChunks) {\n\t\t\tchunkMap.set(chunk.id, chunk);\n\t\t}\n\n\t\tconst results: SearchResult[] = [];\n\t\tfor (const candidate of ranked.slice(0, limit)) {\n\t\t\tconst chunk = chunkMap.get(candidate.id);\n\t\t\tif (chunk) {\n\t\t\t\tresults.push({\n\t\t\t\t\tchunk,\n\t\t\t\t\tscores: candidate.scores,\n\t\t\t\t\trank: candidate.rank,\n\t\t\t\t});\n\t\t\t}\n\t\t}\n\n\t\treturn results;\n\t}\n\n\t/** Get index stats without opening a new connection. */\n\tgetStats(): { files: number; chunks: number } | null {\n\t\tif (!this.indexManager) return null;\n\t\treturn this.indexManager.getStats();\n\t}\n\n\t/** Dispose resources. */\n\tclose(): void {\n\t\tthis.indexManager?.close();\n\t\tthis.indexManager = null;\n\t\tthis.embedder?.dispose();\n\t\tthis.embedder = null;\n\t}\n\n\t// ========================================================================\n\t// Private\n\t// ========================================================================\n\n\tprivate getIndexManager(): IndexManager {\n\t\tif (!this.indexManager) {\n\t\t\tconst config = this.getIndexConfig();\n\t\t\tthis.indexManager = new IndexManager(config);\n\t\t\tthis.indexManager.open();\n\t\t}\n\t\treturn this.indexManager;\n\t}\n\n\tprivate getIndexConfig(): IndexConfig {\n\t\treturn {\n\t\t\tprojectRoot: this.projectRoot,\n\t\t\tindexDir: path.join(this.projectRoot, \".dreb\", \"index\"),\n\t\t\tglobalMemoryDir: path.join(homedir(), \".dreb\", \"memory\"),\n\t\t\tmodelName: DEFAULT_MODEL_NAME,\n\t\t};\n\t}\n\n\tprivate async getOrCreateEmbedder(): Promise<Embedder> {\n\t\tif (!this.embedder) {\n\t\t\tconst config = this.getIndexConfig();\n\t\t\tthis.embedder = new Embedder({\n\t\t\t\tmodelCacheDir: path.join(homedir(), \".dreb\", \"agent\", \"models\"),\n\t\t\t\tmodelName: config.modelName,\n\t\t\t});\n\t\t\tawait this.embedder.initialize();\n\t\t}\n\t\treturn this.embedder;\n\t}\n\n\tprivate async computeVectorScores(\n\t\tdb: SearchDatabase,\n\t\tquery: string,\n\t\tlimit: number,\n\t\t_onProgress?: IndexProgressCallback,\n\t): Promise<Map<number, number>> {\n\t\tconst config = this.getIndexConfig();\n\t\tconst embedder = await this.getOrCreateEmbedder();\n\n\t\t// Embed the query\n\t\tconst queryVector = await embedder.embedQuery(query);\n\n\t\t// Get all stored embeddings\n\t\tconst storedVectors = db.getAllEmbeddings(config.modelName);\n\n\t\tif (storedVectors.size === 0) {\n\t\t\treturn new Map();\n\t\t}\n\n\t\tconst topK = topKSimilar(queryVector, storedVectors, limit);\n\n\t\t// Convert to Map, clamping negative similarities to 0\n\t\tconst scores = new Map<number, number>();\n\t\tfor (const { id, score } of topK) {\n\t\t\tscores.set(id, Math.max(0, score));\n\t\t}\n\t\treturn scores;\n\t}\n}\n\n// ============================================================================\n// Helpers\n// ============================================================================\n\n/** Collect all unique chunk IDs that appear in any metric's results. */\nfunction collectCandidateIds(...scoreMaps: Map<number, number>[]): Set<number> {\n\tconst ids = new Set<number>();\n\tfor (const map of scoreMaps) {\n\t\tfor (const id of map.keys()) {\n\t\t\tids.add(id);\n\t\t}\n\t}\n\treturn ids;\n}\n\n/** Aggregate chunk-level scores to file-level scores (max per file). */\nfunction aggregateFileScores(chunks: StoredChunk[], ...scoreMaps: Map<number, number>[]): Map<number, number> {\n\tconst fileScores = new Map<number, number>();\n\n\tfor (const chunk of chunks) {\n\t\tlet maxScore = 0;\n\t\tfor (const map of scoreMaps) {\n\t\t\tconst s = map.get(chunk.id);\n\t\t\tif (s !== undefined && s > maxScore) maxScore = s;\n\t\t}\n\t\tif (maxScore > 0) {\n\t\t\tconst existing = fileScores.get(chunk.fileId);\n\t\t\tif (existing === undefined || maxScore > existing) {\n\t\t\t\tfileScores.set(chunk.fileId, maxScore);\n\t\t\t}\n\t\t}\n\t}\n\n\treturn fileScores;\n}\n\n/**\n * Compute a dynamic threshold for import graph seeds.\n * Uses the median score — only the top half of files are strong enough seeds.\n * Falls back to 0.1 minimum to avoid accepting near-zero scores.\n */\nfunction computeSeedThreshold(fileScores: Map<number, number>): number {\n\tif (fileScores.size === 0) return 0;\n\tconst sorted = [...fileScores.values()].sort((a, b) => b - a);\n\tconst median = sorted[Math.floor(sorted.length / 2)];\n\treturn Math.max(median, 0.1);\n}\n\n/** Build a map of fileId → chunk IDs for that file. */\nfunction buildFileChunkMap(chunks: StoredChunk[]): Map<number, number[]> {\n\tconst map = new Map<number, number[]>();\n\tfor (const chunk of chunks) {\n\t\tconst existing = map.get(chunk.fileId);\n\t\tif (existing) existing.push(chunk.id);\n\t\telse map.set(chunk.fileId, [chunk.id]);\n\t}\n\treturn map;\n}\n\n/** Common English stopwords to exclude from FTS queries. */\nconst STOPWORDS = new Set([\n\t\"a\",\n\t\"an\",\n\t\"and\",\n\t\"are\",\n\t\"as\",\n\t\"at\",\n\t\"be\",\n\t\"but\",\n\t\"by\",\n\t\"for\",\n\t\"from\",\n\t\"had\",\n\t\"has\",\n\t\"have\",\n\t\"he\",\n\t\"her\",\n\t\"his\",\n\t\"how\",\n\t\"i\",\n\t\"if\",\n\t\"in\",\n\t\"into\",\n\t\"is\",\n\t\"it\",\n\t\"its\",\n\t\"me\",\n\t\"my\",\n\t\"no\",\n\t\"not\",\n\t\"of\",\n\t\"on\",\n\t\"or\",\n\t\"our\",\n\t\"she\",\n\t\"so\",\n\t\"than\",\n\t\"that\",\n\t\"the\",\n\t\"their\",\n\t\"them\",\n\t\"then\",\n\t\"there\",\n\t\"these\",\n\t\"they\",\n\t\"this\",\n\t\"to\",\n\t\"up\",\n\t\"us\",\n\t\"was\",\n\t\"we\",\n\t\"what\",\n\t\"when\",\n\t\"where\",\n\t\"which\",\n\t\"who\",\n\t\"will\",\n\t\"with\",\n\t\"would\",\n\t\"you\",\n\t\"your\",\n]);\n\n/**\n * Sanitize a query string for FTS5 MATCH syntax.\n * FTS5 chokes on certain characters — strip operators and wrap terms.\n *\n * Removes stopwords and uses OR between terms so multi-word queries return\n * partial matches (FTS5's default implicit AND is too restrictive).\n */\nfunction sanitizeFtsQuery(query: string): string {\n\t// Remove FTS5 operators and special chars\n\tconst cleaned = query\n\t\t.replace(/[*\"():^{}[\\]~!@#$%&=+|<>]/g, \" \")\n\t\t.replace(/\\bAND\\b|\\bOR\\b|\\bNOT\\b|\\bNEAR\\b/gi, \" \")\n\t\t.trim();\n\n\t// Split into tokens, remove stopwords, join with OR\n\tconst tokens = cleaned.split(/\\s+/).filter((t) => t.length > 0 && !STOPWORDS.has(t.toLowerCase()));\n\tif (tokens.length === 0) return '\"\"';\n\tif (tokens.length === 1) return tokens[0];\n\treturn tokens.join(\" OR \");\n}\n"]}