@kolisachint/hoocode-agent 0.4.144 → 0.4.145

package/CHANGELOG.md

@@ -1,5 +1,7 @@
 # Changelog
 
+## [0.4.145] - 2026-07-18
+
 ## [0.4.144] - 2026-07-18
 
 ## [0.4.143] - 2026-07-18

package/dist/core/search/context-assembler.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"context-assembler.d.ts","sourceRoot":"","sources":["../../../src/core/search/context-assembler.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAIH,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AASjD,MAAM,WAAW,eAAe;IAC/B,GAAG,EAAE,MAAM,CAAC;IACZ,0DAA0D;IAC1D,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,gBAAgB;IAChC,IAAI,EAAE,MAAM,CAAC;IACb,6EAA6E;IAC7E,YAAY,EAAE,MAAM,CAAC;CACrB;AAMD,wBAAgB,eAAe,CAAC,UAAU,EAAE,SAAS,cAAc,EAAE,EAAE,OAAO,EAAE,eAAe,GAAG,gBAAgB,CAoDjH","sourcesContent":["/**\n * Token-budgeted span expansion (docs/hybrid-retrieval-design.md, step 5 of\n * the shipping order).\n *\n * Retrieval works on chunk ids; only here — after fusion — are line windows\n * read from disk. Every candidate gets a compact `path:start-end [sources]`\n * header; snippets are added top-down until the budget runs out, so the model\n * always sees the full ranked list but never an unbounded dump.\n */\n\nimport { readFileSync } from \"fs\";\nimport path from \"path\";\nimport type { FusedCandidate } from \"./types.js\";\n\n/** Rough chars-per-token for budgeting (index-time uses the same heuristic). */\nconst CHARS_PER_TOKEN = 4;\nconst DEFAULT_TOKEN_BUDGET = 2000;\n/** Snippet caps keep one giant chunk from eating the whole budget. */\nconst MAX_SNIPPET_LINES = 20;\nconst MAX_SNIPPET_LINE_CHARS = 200;\n\nexport interface AssembleOptions {\n\tcwd: string;\n\t/** Approximate token budget for the whole result text. */\n\ttokenBudget?: number;\n}\n\nexport interface AssembledContext {\n\ttext: string;\n\t/** How many candidates got an inline snippet (the rest are bare headers). */\n\tsnippetCount: number;\n}\n\nfunction sourcesLabel(candidate: FusedCandidate): string {\n\treturn Object.keys(candidate.ranks).sort().join(\"+\");\n}\n\nexport function assembleContext(candidates: readonly FusedCandidate[], options: AssembleOptions): AssembledContext {\n\tconst budgetChars = (options.tokenBudget ?? DEFAULT_TOKEN_BUDGET) * CHARS_PER_TOKEN;\n\tconst fileCache = new Map<string, string[] | undefined>();\n\n\tconst readLines = (rel: string): string[] | undefined => {\n\t\tif (!fileCache.has(rel)) {\n\t\t\ttry {\n\t\t\t\tconst content = readFileSync(path.resolve(options.cwd, rel), \"utf-8\");\n\t\t\t\tfileCache.set(rel, content.replace(/\\r\\n/g, \"\\n\").replace(/\\r/g, \"\\n\").split(\"\\n\"));\n\t\t\t} catch {\n\t\t\t\tfileCache.set(rel, undefined);\n\t\t\t}\n\t\t}\n\t\treturn fileCache.get(rel);\n\t};\n\n\tconst sections: string[] = [];\n\tlet usedChars = 0;\n\tlet snippetCount = 0;\n\tlet snippetsExhausted = false;\n\n\tfor (const candidate of candidates) {\n\t\tconst lines = readLines(candidate.path);\n\t\t// Clamp the span to the file as it exists now (fallback spans may\n\t\t// overshoot; the file may have changed since indexing).\n\t\tconst start = Math.max(1, candidate.startLine);\n\t\tconst end = lines ? Math.min(candidate.endLine, lines.length) : candidate.endLine;\n\t\tconst header = `${candidate.path}:${start}-${end} [${sourcesLabel(candidate)}]`;\n\t\tusedChars += header.length + 1;\n\n\t\tif (!snippetsExhausted && lines && end >= start) {\n\t\t\tconst snippetEnd = Math.min(end, start + MAX_SNIPPET_LINES - 1);\n\t\t\tconst snippetLines = lines\n\t\t\t\t.slice(start - 1, snippetEnd)\n\t\t\t\t.map(\n\t\t\t\t\t(text, i) =>\n\t\t\t\t\t\t`  ${start + i}: ${text.length > MAX_SNIPPET_LINE_CHARS ? `${text.slice(0, MAX_SNIPPET_LINE_CHARS)}…` : text}`,\n\t\t\t\t);\n\t\t\tconst snippet = snippetLines.join(\"\\n\");\n\t\t\tif (usedChars + snippet.length <= budgetChars) {\n\t\t\t\tsections.push(`${header}\\n${snippet}`);\n\t\t\t\tusedChars += snippet.length + 1;\n\t\t\t\tsnippetCount++;\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\t// Budget hit: stop expanding, keep listing bare headers.\n\t\t\tsnippetsExhausted = true;\n\t\t}\n\t\tsections.push(header);\n\t}\n\n\treturn { text: sections.join(\"\\n\\n\"), snippetCount };\n}\n"]}
+{"version":3,"file":"context-assembler.d.ts","sourceRoot":"","sources":["../../../src/core/search/context-assembler.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAIH,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAajD,MAAM,WAAW,eAAe;IAC/B,GAAG,EAAE,MAAM,CAAC;IACZ,0DAA0D;IAC1D,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,gBAAgB;IAChC,IAAI,EAAE,MAAM,CAAC;IACb,6EAA6E;IAC7E,YAAY,EAAE,MAAM,CAAC;CACrB;AAMD,wBAAgB,eAAe,CAAC,UAAU,EAAE,SAAS,cAAc,EAAE,EAAE,OAAO,EAAE,eAAe,GAAG,gBAAgB,CAyDjH","sourcesContent":["/**\n * Token-budgeted span expansion (docs/hybrid-retrieval-design.md, step 5 of\n * the shipping order).\n *\n * Retrieval works on chunk ids; only here — after fusion — are line windows\n * read from disk. Every candidate gets a compact `path:start-end [sources]`\n * header; snippets are added top-down until the budget runs out, so the model\n * always sees the full ranked list but never an unbounded dump.\n */\n\nimport { readFileSync } from \"fs\";\nimport path from \"path\";\nimport type { FusedCandidate } from \"./types.js\";\n\n/** Rough chars-per-token for budgeting (index-time uses the same heuristic). */\nconst CHARS_PER_TOKEN = 4;\nconst DEFAULT_TOKEN_BUDGET = 2000;\n/** Snippet caps keep one giant chunk from eating the whole budget. Results\n *  past the top few get a shallower snippet: rank carries most of the value,\n *  and full-depth snippets for every result roughly doubles the token cost. */\nconst MAX_SNIPPET_LINES = 20;\nconst TOP_FULL_SNIPPETS = 3;\nconst TAIL_SNIPPET_LINES = 8;\nconst MAX_SNIPPET_LINE_CHARS = 200;\n\nexport interface AssembleOptions {\n\tcwd: string;\n\t/** Approximate token budget for the whole result text. */\n\ttokenBudget?: number;\n}\n\nexport interface AssembledContext {\n\ttext: string;\n\t/** How many candidates got an inline snippet (the rest are bare headers). */\n\tsnippetCount: number;\n}\n\nfunction sourcesLabel(candidate: FusedCandidate): string {\n\treturn Object.keys(candidate.ranks).sort().join(\"+\");\n}\n\nexport function assembleContext(candidates: readonly FusedCandidate[], options: AssembleOptions): AssembledContext {\n\tconst budgetChars = (options.tokenBudget ?? DEFAULT_TOKEN_BUDGET) * CHARS_PER_TOKEN;\n\tconst fileCache = new Map<string, string[] | undefined>();\n\n\tconst readLines = (rel: string): string[] | undefined => {\n\t\tif (!fileCache.has(rel)) {\n\t\t\ttry {\n\t\t\t\tconst content = readFileSync(path.resolve(options.cwd, rel), \"utf-8\");\n\t\t\t\tfileCache.set(rel, content.replace(/\\r\\n/g, \"\\n\").replace(/\\r/g, \"\\n\").split(\"\\n\"));\n\t\t\t} catch {\n\t\t\t\tfileCache.set(rel, undefined);\n\t\t\t}\n\t\t}\n\t\treturn fileCache.get(rel);\n\t};\n\n\tconst sections: string[] = [];\n\tlet usedChars = 0;\n\tlet snippetCount = 0;\n\tlet snippetsExhausted = false;\n\n\tfor (const candidate of candidates) {\n\t\tconst lines = readLines(candidate.path);\n\t\t// Clamp the span to the file as it exists now (fallback spans may\n\t\t// overshoot; the file may have changed since indexing).\n\t\tconst start = Math.max(1, candidate.startLine);\n\t\tconst end = lines ? Math.min(candidate.endLine, lines.length) : candidate.endLine;\n\t\tconst header = `${candidate.path}:${start}-${end} [${sourcesLabel(candidate)}]`;\n\t\tusedChars += header.length + 1;\n\n\t\tif (!snippetsExhausted && lines && end >= start) {\n\t\t\tconst depth = snippetCount < TOP_FULL_SNIPPETS ? MAX_SNIPPET_LINES : TAIL_SNIPPET_LINES;\n\t\t\tconst snippetEnd = Math.min(end, start + depth - 1);\n\t\t\tconst rawLines = lines.slice(start - 1, snippetEnd);\n\t\t\t// Chunk spans often end on a blank line (trailing-newline artifact);\n\t\t\t// trailing blanks carry no signal, so drop them.\n\t\t\twhile (rawLines.length > 0 && rawLines[rawLines.length - 1].trim() === \"\") rawLines.pop();\n\t\t\tconst snippetLines = rawLines.map(\n\t\t\t\t(text, i) =>\n\t\t\t\t\t`  ${start + i}: ${text.length > MAX_SNIPPET_LINE_CHARS ? `${text.slice(0, MAX_SNIPPET_LINE_CHARS)}…` : text}`,\n\t\t\t);\n\t\t\tconst snippet = snippetLines.join(\"\\n\");\n\t\t\tif (snippet) {\n\t\t\t\tif (usedChars + snippet.length <= budgetChars) {\n\t\t\t\t\tsections.push(`${header}\\n${snippet}`);\n\t\t\t\t\tusedChars += snippet.length + 1;\n\t\t\t\t\tsnippetCount++;\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\t// Budget hit: stop expanding, keep listing bare headers.\n\t\t\t\tsnippetsExhausted = true;\n\t\t\t}\n\t\t}\n\t\tsections.push(header);\n\t}\n\n\treturn { text: sections.join(\"\\n\\n\"), snippetCount };\n}\n"]}

package/dist/core/search/context-assembler.js

@@ -12,8 +12,12 @@ import path from "path";
 /** Rough chars-per-token for budgeting (index-time uses the same heuristic). */
 const CHARS_PER_TOKEN = 4;
 const DEFAULT_TOKEN_BUDGET = 2000;
-/** Snippet caps keep one giant chunk from eating the whole budget. */
+/** Snippet caps keep one giant chunk from eating the whole budget. Results
+ *  past the top few get a shallower snippet: rank carries most of the value,
+ *  and full-depth snippets for every result roughly doubles the token cost. */
 const MAX_SNIPPET_LINES = 20;
+const TOP_FULL_SNIPPETS = 3;
+const TAIL_SNIPPET_LINES = 8;
 const MAX_SNIPPET_LINE_CHARS = 200;
 function sourcesLabel(candidate) {
     return Object.keys(candidate.ranks).sort().join("+");
@@ -46,19 +50,25 @@ export function assembleContext(candidates, options) {
         const header = `${candidate.path}:${start}-${end} [${sourcesLabel(candidate)}]`;
         usedChars += header.length + 1;
         if (!snippetsExhausted && lines && end >= start) {
-            const snippetEnd = Math.min(end, start + MAX_SNIPPET_LINES - 1);
-            const snippetLines = lines
-                .slice(start - 1, snippetEnd)
-                .map((text, i) => `  ${start + i}: ${text.length > MAX_SNIPPET_LINE_CHARS ? `${text.slice(0, MAX_SNIPPET_LINE_CHARS)}…` : text}`);
+            const depth = snippetCount < TOP_FULL_SNIPPETS ? MAX_SNIPPET_LINES : TAIL_SNIPPET_LINES;
+            const snippetEnd = Math.min(end, start + depth - 1);
+            const rawLines = lines.slice(start - 1, snippetEnd);
+            // Chunk spans often end on a blank line (trailing-newline artifact);
+            // trailing blanks carry no signal, so drop them.
+            while (rawLines.length > 0 && rawLines[rawLines.length - 1].trim() === "")
+                rawLines.pop();
+            const snippetLines = rawLines.map((text, i) => `  ${start + i}: ${text.length > MAX_SNIPPET_LINE_CHARS ? `${text.slice(0, MAX_SNIPPET_LINE_CHARS)}…` : text}`);
             const snippet = snippetLines.join("\n");
-            if (usedChars + snippet.length <= budgetChars) {
-                sections.push(`${header}\n${snippet}`);
-                usedChars += snippet.length + 1;
-                snippetCount++;
-                continue;
+            if (snippet) {
+                if (usedChars + snippet.length <= budgetChars) {
+                    sections.push(`${header}\n${snippet}`);
+                    usedChars += snippet.length + 1;
+                    snippetCount++;
+                    continue;
+                }
+                // Budget hit: stop expanding, keep listing bare headers.
+                snippetsExhausted = true;
             }
-            // Budget hit: stop expanding, keep listing bare headers.
-            snippetsExhausted = true;
         }
         sections.push(header);
     }

package/dist/core/search/context-assembler.js.map

@@ -1 +1 @@
-{"version":3,"file":"context-assembler.js","sourceRoot":"","sources":["../../../src/core/search/context-assembler.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;AAClC,OAAO,IAAI,MAAM,MAAM,CAAC;AAGxB,gFAAgF;AAChF,MAAM,eAAe,GAAG,CAAC,CAAC;AAC1B,MAAM,oBAAoB,GAAG,IAAI,CAAC;AAClC,sEAAsE;AACtE,MAAM,iBAAiB,GAAG,EAAE,CAAC;AAC7B,MAAM,sBAAsB,GAAG,GAAG,CAAC;AAcnC,SAAS,YAAY,CAAC,SAAyB,EAAU;IACxD,OAAO,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;AAAA,CACrD;AAED,MAAM,UAAU,eAAe,CAAC,UAAqC,EAAE,OAAwB,EAAoB;IAClH,MAAM,WAAW,GAAG,CAAC,OAAO,CAAC,WAAW,IAAI,oBAAoB,CAAC,GAAG,eAAe,CAAC;IACpF,MAAM,SAAS,GAAG,IAAI,GAAG,EAAgC,CAAC;IAE1D,MAAM,SAAS,GAAG,CAAC,GAAW,EAAwB,EAAE,CAAC;QACxD,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;YACzB,IAAI,CAAC;gBACJ,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,OAAO,CAAC,CAAC;gBACtE,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;YACrF,CAAC;YAAC,MAAM,CAAC;gBACR,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE,SAAS,CAAC,CAAC;YAC/B,CAAC;QACF,CAAC;QACD,OAAO,SAAS,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAAA,CAC1B,CAAC;IAEF,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,iBAAiB,GAAG,KAAK,CAAC;IAE9B,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,MAAM,KAAK,GAAG,SAAS,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;QACxC,kEAAkE;QAClE,wDAAwD;QACxD,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,CAAC,SAAS,CAAC,CAAC;QAC/C,MAAM,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,SAAS,CAAC,OAAO,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,OAAO,CAAC;QAClF,MAAM,MAAM,GAAG,GAAG,SAAS,CAAC,IAAI,IAAI,KAAK,IAAI,GAAG,KAAK,YAAY,CAAC,SAAS,CAAC,GAAG,CAAC;QAChF,SAAS,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC;QAE/B,IAAI,CAAC,iBAAiB,IAAI,KAAK,IAAI,GAAG,IAAI,KAAK,EAAE,CAAC;YACjD,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,GAAG,iBAAiB,GAAG,CAAC,CAAC,CAAC;YAChE,MAAM,YAAY,GAAG,KAAK;iBACxB,KAAK,CAAC,KAAK,GAAG,CAAC,EAAE,UAAU,CAAC;iBAC5B,GAAG,CACH,CAAC,IAAI,EAAE,CAAC,EAAE,EAAE,CACX,KAAK,KAAK,GAAG,CAAC,KAAK,IAAI,CAAC,MAAM,GAAG,sBAAsB,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,sBAAsB,CAAC,KAAG,CAAC,CAAC,CAAC,IAAI,EAAE,CAC/G,CAAC;YACH,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACxC,IAAI,SAAS,GAAG,OAAO,CAAC,MAAM,IAAI,WAAW,EAAE,CAAC;gBAC/C,QAAQ,CAAC,IAAI,CAAC,GAAG,MAAM,KAAK,OAAO,EAAE,CAAC,CAAC;gBACvC,SAAS,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC;gBAChC,YAAY,EAAE,CAAC;gBACf,SAAS;YACV,CAAC;YACD,yDAAyD;YACzD,iBAAiB,GAAG,IAAI,CAAC;QAC1B,CAAC;QACD,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACvB,CAAC;IAED,OAAO,EAAE,IAAI,EAAE,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,YAAY,EAAE,CAAC;AAAA,CACrD","sourcesContent":["/**\n * Token-budgeted span expansion (docs/hybrid-retrieval-design.md, step 5 of\n * the shipping order).\n *\n * Retrieval works on chunk ids; only here — after fusion — are line windows\n * read from disk. Every candidate gets a compact `path:start-end [sources]`\n * header; snippets are added top-down until the budget runs out, so the model\n * always sees the full ranked list but never an unbounded dump.\n */\n\nimport { readFileSync } from \"fs\";\nimport path from \"path\";\nimport type { FusedCandidate } from \"./types.js\";\n\n/** Rough chars-per-token for budgeting (index-time uses the same heuristic). */\nconst CHARS_PER_TOKEN = 4;\nconst DEFAULT_TOKEN_BUDGET = 2000;\n/** Snippet caps keep one giant chunk from eating the whole budget. */\nconst MAX_SNIPPET_LINES = 20;\nconst MAX_SNIPPET_LINE_CHARS = 200;\n\nexport interface AssembleOptions {\n\tcwd: string;\n\t/** Approximate token budget for the whole result text. */\n\ttokenBudget?: number;\n}\n\nexport interface AssembledContext {\n\ttext: string;\n\t/** How many candidates got an inline snippet (the rest are bare headers). */\n\tsnippetCount: number;\n}\n\nfunction sourcesLabel(candidate: FusedCandidate): string {\n\treturn Object.keys(candidate.ranks).sort().join(\"+\");\n}\n\nexport function assembleContext(candidates: readonly FusedCandidate[], options: AssembleOptions): AssembledContext {\n\tconst budgetChars = (options.tokenBudget ?? DEFAULT_TOKEN_BUDGET) * CHARS_PER_TOKEN;\n\tconst fileCache = new Map<string, string[] | undefined>();\n\n\tconst readLines = (rel: string): string[] | undefined => {\n\t\tif (!fileCache.has(rel)) {\n\t\t\ttry {\n\t\t\t\tconst content = readFileSync(path.resolve(options.cwd, rel), \"utf-8\");\n\t\t\t\tfileCache.set(rel, content.replace(/\\r\\n/g, \"\\n\").replace(/\\r/g, \"\\n\").split(\"\\n\"));\n\t\t\t} catch {\n\t\t\t\tfileCache.set(rel, undefined);\n\t\t\t}\n\t\t}\n\t\treturn fileCache.get(rel);\n\t};\n\n\tconst sections: string[] = [];\n\tlet usedChars = 0;\n\tlet snippetCount = 0;\n\tlet snippetsExhausted = false;\n\n\tfor (const candidate of candidates) {\n\t\tconst lines = readLines(candidate.path);\n\t\t// Clamp the span to the file as it exists now (fallback spans may\n\t\t// overshoot; the file may have changed since indexing).\n\t\tconst start = Math.max(1, candidate.startLine);\n\t\tconst end = lines ? Math.min(candidate.endLine, lines.length) : candidate.endLine;\n\t\tconst header = `${candidate.path}:${start}-${end} [${sourcesLabel(candidate)}]`;\n\t\tusedChars += header.length + 1;\n\n\t\tif (!snippetsExhausted && lines && end >= start) {\n\t\t\tconst snippetEnd = Math.min(end, start + MAX_SNIPPET_LINES - 1);\n\t\t\tconst snippetLines = lines\n\t\t\t\t.slice(start - 1, snippetEnd)\n\t\t\t\t.map(\n\t\t\t\t\t(text, i) =>\n\t\t\t\t\t\t`  ${start + i}: ${text.length > MAX_SNIPPET_LINE_CHARS ? `${text.slice(0, MAX_SNIPPET_LINE_CHARS)}…` : text}`,\n\t\t\t\t);\n\t\t\tconst snippet = snippetLines.join(\"\\n\");\n\t\t\tif (usedChars + snippet.length <= budgetChars) {\n\t\t\t\tsections.push(`${header}\\n${snippet}`);\n\t\t\t\tusedChars += snippet.length + 1;\n\t\t\t\tsnippetCount++;\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\t// Budget hit: stop expanding, keep listing bare headers.\n\t\t\tsnippetsExhausted = true;\n\t\t}\n\t\tsections.push(header);\n\t}\n\n\treturn { text: sections.join(\"\\n\\n\"), snippetCount };\n}\n"]}
+{"version":3,"file":"context-assembler.js","sourceRoot":"","sources":["../../../src/core/search/context-assembler.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;AAClC,OAAO,IAAI,MAAM,MAAM,CAAC;AAGxB,gFAAgF;AAChF,MAAM,eAAe,GAAG,CAAC,CAAC;AAC1B,MAAM,oBAAoB,GAAG,IAAI,CAAC;AAClC;;+EAE+E;AAC/E,MAAM,iBAAiB,GAAG,EAAE,CAAC;AAC7B,MAAM,iBAAiB,GAAG,CAAC,CAAC;AAC5B,MAAM,kBAAkB,GAAG,CAAC,CAAC;AAC7B,MAAM,sBAAsB,GAAG,GAAG,CAAC;AAcnC,SAAS,YAAY,CAAC,SAAyB,EAAU;IACxD,OAAO,MAAM,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;AAAA,CACrD;AAED,MAAM,UAAU,eAAe,CAAC,UAAqC,EAAE,OAAwB,EAAoB;IAClH,MAAM,WAAW,GAAG,CAAC,OAAO,CAAC,WAAW,IAAI,oBAAoB,CAAC,GAAG,eAAe,CAAC;IACpF,MAAM,SAAS,GAAG,IAAI,GAAG,EAAgC,CAAC;IAE1D,MAAM,SAAS,GAAG,CAAC,GAAW,EAAwB,EAAE,CAAC;QACxD,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;YACzB,IAAI,CAAC;gBACJ,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,OAAO,CAAC,CAAC;gBACtE,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE,OAAO,CAAC,OAAO,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;YACrF,CAAC;YAAC,MAAM,CAAC;gBACR,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE,SAAS,CAAC,CAAC;YAC/B,CAAC;QACF,CAAC;QACD,OAAO,SAAS,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAAA,CAC1B,CAAC;IAEF,MAAM,QAAQ,GAAa,EAAE,CAAC;IAC9B,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,iBAAiB,GAAG,KAAK,CAAC;IAE9B,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,MAAM,KAAK,GAAG,SAAS,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC;QACxC,kEAAkE;QAClE,wDAAwD;QACxD,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,SAAS,CAAC,SAAS,CAAC,CAAC;QAC/C,MAAM,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,SAAS,CAAC,OAAO,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,OAAO,CAAC;QAClF,MAAM,MAAM,GAAG,GAAG,SAAS,CAAC,IAAI,IAAI,KAAK,IAAI,GAAG,KAAK,YAAY,CAAC,SAAS,CAAC,GAAG,CAAC;QAChF,SAAS,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC;QAE/B,IAAI,CAAC,iBAAiB,IAAI,KAAK,IAAI,GAAG,IAAI,KAAK,EAAE,CAAC;YACjD,MAAM,KAAK,GAAG,YAAY,GAAG,iBAAiB,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,kBAAkB,CAAC;YACxF,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,KAAK,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC;YACpD,MAAM,QAAQ,GAAG,KAAK,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,EAAE,UAAU,CAAC,CAAC;YACpD,qEAAqE;YACrE,iDAAiD;YACjD,OAAO,QAAQ,CAAC,MAAM,GAAG,CAAC,IAAI,QAAQ,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE;gBAAE,QAAQ,CAAC,GAAG,EAAE,CAAC;YAC1F,MAAM,YAAY,GAAG,QAAQ,CAAC,GAAG,CAChC,CAAC,IAAI,EAAE,CAAC,EAAE,EAAE,CACX,KAAK,KAAK,GAAG,CAAC,KAAK,IAAI,CAAC,MAAM,GAAG,sBAAsB,CAAC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,sBAAsB,CAAC,KAAG,CAAC,CAAC,CAAC,IAAI,EAAE,CAC/G,CAAC;YACF,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACxC,IAAI,OAAO,EAAE,CAAC;gBACb,IAAI,SAAS,GAAG,OAAO,CAAC,MAAM,IAAI,WAAW,EAAE,CAAC;oBAC/C,QAAQ,CAAC,IAAI,CAAC,GAAG,MAAM,KAAK,OAAO,EAAE,CAAC,CAAC;oBACvC,SAAS,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC;oBAChC,YAAY,EAAE,CAAC;oBACf,SAAS;gBACV,CAAC;gBACD,yDAAyD;gBACzD,iBAAiB,GAAG,IAAI,CAAC;YAC1B,CAAC;QACF,CAAC;QACD,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACvB,CAAC;IAED,OAAO,EAAE,IAAI,EAAE,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,YAAY,EAAE,CAAC;AAAA,CACrD","sourcesContent":["/**\n * Token-budgeted span expansion (docs/hybrid-retrieval-design.md, step 5 of\n * the shipping order).\n *\n * Retrieval works on chunk ids; only here — after fusion — are line windows\n * read from disk. Every candidate gets a compact `path:start-end [sources]`\n * header; snippets are added top-down until the budget runs out, so the model\n * always sees the full ranked list but never an unbounded dump.\n */\n\nimport { readFileSync } from \"fs\";\nimport path from \"path\";\nimport type { FusedCandidate } from \"./types.js\";\n\n/** Rough chars-per-token for budgeting (index-time uses the same heuristic). */\nconst CHARS_PER_TOKEN = 4;\nconst DEFAULT_TOKEN_BUDGET = 2000;\n/** Snippet caps keep one giant chunk from eating the whole budget. Results\n *  past the top few get a shallower snippet: rank carries most of the value,\n *  and full-depth snippets for every result roughly doubles the token cost. */\nconst MAX_SNIPPET_LINES = 20;\nconst TOP_FULL_SNIPPETS = 3;\nconst TAIL_SNIPPET_LINES = 8;\nconst MAX_SNIPPET_LINE_CHARS = 200;\n\nexport interface AssembleOptions {\n\tcwd: string;\n\t/** Approximate token budget for the whole result text. */\n\ttokenBudget?: number;\n}\n\nexport interface AssembledContext {\n\ttext: string;\n\t/** How many candidates got an inline snippet (the rest are bare headers). */\n\tsnippetCount: number;\n}\n\nfunction sourcesLabel(candidate: FusedCandidate): string {\n\treturn Object.keys(candidate.ranks).sort().join(\"+\");\n}\n\nexport function assembleContext(candidates: readonly FusedCandidate[], options: AssembleOptions): AssembledContext {\n\tconst budgetChars = (options.tokenBudget ?? DEFAULT_TOKEN_BUDGET) * CHARS_PER_TOKEN;\n\tconst fileCache = new Map<string, string[] | undefined>();\n\n\tconst readLines = (rel: string): string[] | undefined => {\n\t\tif (!fileCache.has(rel)) {\n\t\t\ttry {\n\t\t\t\tconst content = readFileSync(path.resolve(options.cwd, rel), \"utf-8\");\n\t\t\t\tfileCache.set(rel, content.replace(/\\r\\n/g, \"\\n\").replace(/\\r/g, \"\\n\").split(\"\\n\"));\n\t\t\t} catch {\n\t\t\t\tfileCache.set(rel, undefined);\n\t\t\t}\n\t\t}\n\t\treturn fileCache.get(rel);\n\t};\n\n\tconst sections: string[] = [];\n\tlet usedChars = 0;\n\tlet snippetCount = 0;\n\tlet snippetsExhausted = false;\n\n\tfor (const candidate of candidates) {\n\t\tconst lines = readLines(candidate.path);\n\t\t// Clamp the span to the file as it exists now (fallback spans may\n\t\t// overshoot; the file may have changed since indexing).\n\t\tconst start = Math.max(1, candidate.startLine);\n\t\tconst end = lines ? Math.min(candidate.endLine, lines.length) : candidate.endLine;\n\t\tconst header = `${candidate.path}:${start}-${end} [${sourcesLabel(candidate)}]`;\n\t\tusedChars += header.length + 1;\n\n\t\tif (!snippetsExhausted && lines && end >= start) {\n\t\t\tconst depth = snippetCount < TOP_FULL_SNIPPETS ? MAX_SNIPPET_LINES : TAIL_SNIPPET_LINES;\n\t\t\tconst snippetEnd = Math.min(end, start + depth - 1);\n\t\t\tconst rawLines = lines.slice(start - 1, snippetEnd);\n\t\t\t// Chunk spans often end on a blank line (trailing-newline artifact);\n\t\t\t// trailing blanks carry no signal, so drop them.\n\t\t\twhile (rawLines.length > 0 && rawLines[rawLines.length - 1].trim() === \"\") rawLines.pop();\n\t\t\tconst snippetLines = rawLines.map(\n\t\t\t\t(text, i) =>\n\t\t\t\t\t`  ${start + i}: ${text.length > MAX_SNIPPET_LINE_CHARS ? `${text.slice(0, MAX_SNIPPET_LINE_CHARS)}…` : text}`,\n\t\t\t);\n\t\t\tconst snippet = snippetLines.join(\"\\n\");\n\t\t\tif (snippet) {\n\t\t\t\tif (usedChars + snippet.length <= budgetChars) {\n\t\t\t\t\tsections.push(`${header}\\n${snippet}`);\n\t\t\t\t\tusedChars += snippet.length + 1;\n\t\t\t\t\tsnippetCount++;\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\t// Budget hit: stop expanding, keep listing bare headers.\n\t\t\t\tsnippetsExhausted = true;\n\t\t\t}\n\t\t}\n\t\tsections.push(header);\n\t}\n\n\treturn { text: sections.join(\"\\n\\n\"), snippetCount };\n}\n"]}

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

@@ -1 +1 @@
-{"version":3,"file":"hybrid-search.d.ts","sourceRoot":"","sources":["../../../src/core/search/hybrid-search.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,mCAAmC,CAAC;AAQ1E,OAAO,KAAK,EAAiB,cAAc,EAAa,kBAAkB,EAAE,UAAU,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAcxH,MAAM,WAAW,eAAe;IAC/B,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,UAAU,CAAC;IAClB,kDAAkD;IAClD,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,yCAAyC;IACzC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,wFAAwF;IACxF,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,wEAAwE;IACxE,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,OAAO,CAAC,EAAE,gBAAgB,CAAC;IAC3B,MAAM,CAAC,EAAE,WAAW,CAAC;CACrB;AAED,MAAM,WAAW,cAAc;IAC9B,UAAU,EAAE,cAAc,EAAE,CAAC;IAC7B,YAAY,EAAE,kBAAkB,CAAC;IACjC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,UAAU,EAAE,WAAW,CAAC,YAAY,CAAC,CAAC;IACtC,UAAU,EAAE,WAAW,CAAC,YAAY,CAAC,CAAC;IACtC,uDAAuD;IACvD,QAAQ,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAC3C,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,WAAW,CAAC,QAAQ,CAAC,CAAC;CAC/B;AAED,MAAM,WAAW,gBAAiB,SAAQ,eAAe;IACxD,oDAAoD;IACpD,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,eAAe;IAC/B,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,kBAAkB,CAAC;IACjC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,WAAW,EAAE,MAAM,CAAC;IACpB,uDAAuD;IACvD,QAAQ,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;CAC3C;AAYD,wBAAsB,kBAAkB,CAAC,OAAO,EAAE,eAAe,GAAG,OAAO,CAAC,cAAc,CAAC,CAwG1F;AAED,wBAAsB,SAAS,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAAC,eAAe,CAAC,CAyBnF","sourcesContent":["/**\n * Hybrid search orchestrator: resolve mode, run retrievers in parallel, fuse\n * by rank, expand within budget, trace everything\n * (docs/hybrid-retrieval-design.md).\n *\n * Single-retriever modes flow through the same pipeline — rrfFuse over one\n * list preserves its order — so lexical, semantic, and hybrid all produce the\n * same result shape and the same trace record.\n *\n * `retrieveCandidates` is the candidate-level core (also used by the eval\n * harness, which needs forced modes, a configurable `k`, and no trace\n * pollution); `runSearch` wraps it with span expansion and tracing for the\n * tool.\n */\n\nimport type { EmbsearchService } from \"../embsearch/embsearch-service.js\";\nimport { adaptGrepHits, type ChunkLookup } from \"./adapter.js\";\nimport { assembleContext } from \"./context-assembler.js\";\nimport { runLexicalRetriever } from \"./lexical-retriever.js\";\nimport { resolveSearchMode } from \"./mode.js\";\nimport { rerankCandidates } from \"./rerank.js\";\nimport { DEFAULT_RRF_K, rrfFuse } from \"./rrf.js\";\nimport { writeSearchTrace } from \"./trace.js\";\nimport type { CandidateSpan, FusedCandidate, RankedHit, ResolvedSearchMode, SearchMode, SearchTrace } from \"./types.js\";\n\n/** Raw grep line-hits fetched per query (pre-collapse). */\nconst LEXICAL_MATCH_LIMIT = 200;\n/** Adapted lexical candidates entering fusion. The eval gate showed the\n *  uncapped lexical tail diluting hybrid below plain semantic: lexical\n *  precision is front-loaded by the adapter's term-evidence ranking, while\n *  RRF weighs a rank-30 lexical candidate like a rank-30 embedding hit. */\nconst LEXICAL_FUSION_CAP = 20;\n/** Embedding hits fetched per query — deep enough for fusion to matter. */\nconst EMBED_TOP_K = 50;\n/** Fused candidates kept for reranking / final slicing. */\nconst FUSED_WINDOW = 50;\n\nexport interface RetrieveOptions {\n\tcwd: string;\n\tquery: string;\n\tmode?: SearchMode;\n\t/** Optional glob filter applied to file paths. */\n\tglob?: string;\n\t/** Maximum fused candidates returned. */\n\tlimit?: number;\n\t/** RRF constant override (eval harness sweeps this). Default: {@link DEFAULT_RRF_K}. */\n\trrfK?: number;\n\t/** Rerank the fused top-50 before slicing to `limit`. Default: true. */\n\trerank?: boolean;\n\tservice?: EmbsearchService;\n\tsignal?: AbortSignal;\n}\n\nexport interface RetrieveResult {\n\tcandidates: FusedCandidate[];\n\tresolvedMode: ResolvedSearchMode;\n\tdegradedReason?: string;\n\tindexPhase: SearchTrace[\"indexPhase\"];\n\tretrievers: SearchTrace[\"retrievers\"];\n\t/** Set while the embedding index is still building. */\n\tindexing?: { done: number; total: number };\n\trrfK: number;\n\trerank?: SearchTrace[\"rerank\"];\n}\n\nexport interface RunSearchOptions extends RetrieveOptions {\n\t/** Approximate token budget for the result text. */\n\ttokenBudget?: number;\n}\n\nexport interface RunSearchResult {\n\ttext: string;\n\tresolvedMode: ResolvedSearchMode;\n\tdegradedReason?: string;\n\tresultCount: number;\n\t/** Set while the embedding index is still building. */\n\tindexing?: { done: number; total: number };\n}\n\nfunction normalizeSearchGlob(glob: string | undefined): string | undefined {\n\tif (!glob) return undefined;\n\t// Match fd/rg semantics: a slash-containing glob is anchored anywhere in\n\t// the tree, so prepend \"**/\" unless it already starts with a slash or \"**/\".\n\tif (glob.includes(\"/\") && !glob.startsWith(\"/\") && !glob.startsWith(\"**/\")) {\n\t\treturn `**/${glob}`;\n\t}\n\treturn glob;\n}\n\nexport async function retrieveCandidates(options: RetrieveOptions): Promise<RetrieveResult> {\n\tconst { cwd, query, service, signal } = options;\n\tconst glob = normalizeSearchGlob(options.glob);\n\tconst requestedMode = options.mode ?? \"auto\";\n\tconst limit = Math.max(1, options.limit ?? 10);\n\tconst rrfK = options.rrfK ?? DEFAULT_RRF_K;\n\n\tconst state = service?.getState();\n\tconst embedAvailable = service?.isAvailable() ?? false;\n\tconst embedUnavailableReason =\n\t\tstate === undefined\n\t\t\t? \"semantic index is not enabled\"\n\t\t\t: state.phase === \"unavailable\" || state.phase === \"skipped\"\n\t\t\t\t? state.reason\n\t\t\t\t: state.phase === \"idle\"\n\t\t\t\t\t? \"semantic index has not started\"\n\t\t\t\t\t: undefined;\n\n\tconst resolution = resolveSearchMode(query, requestedMode, embedAvailable, embedUnavailableReason);\n\tconst mode = resolution.mode;\n\n\t// Map lexical hits onto indexed chunk ids whenever the sidecar is usable,\n\t// even in lexical-only mode, so identities line up across modes.\n\tconst lookupChunk: ChunkLookup | undefined = embedAvailable\n\t\t? (rel, line) => service!.findEnclosingChunk(rel, line)\n\t\t: undefined;\n\n\tconst spans = new Map<string, CandidateSpan>();\n\tconst lists: RankedHit[][] = [];\n\tconst retrieverStats: SearchTrace[\"retrievers\"] = {};\n\tconst errors: Error[] = [];\n\n\tconst runLexical = async (): Promise<void> => {\n\t\tconst startedMs = Date.now();\n\t\ttry {\n\t\t\tconst lineHits = await runLexicalRetriever({ cwd, query, limit: LEXICAL_MATCH_LIMIT, glob, signal });\n\t\t\tconst adapted = adaptGrepHits(lineHits, lookupChunk);\n\t\t\t// In single-retriever lexical mode the full list is the result; in\n\t\t\t// hybrid, only the front-loaded head is trustworthy enough to vote.\n\t\t\tconst hits = mode === \"hybrid\" ? adapted.hits.slice(0, LEXICAL_FUSION_CAP) : adapted.hits;\n\t\t\tfor (const [id, span] of adapted.spans) if (!spans.has(id)) spans.set(id, span);\n\t\t\tlists.push(hits);\n\t\t\tretrieverStats.grep = { latencyMs: Date.now() - startedMs, hitCount: hits.length };\n\t\t} catch (e) {\n\t\t\terrors.push(e instanceof Error ? e : new Error(String(e)));\n\t\t\tretrieverStats.grep = { latencyMs: Date.now() - startedMs, hitCount: 0 };\n\t\t}\n\t};\n\n\tconst runEmbed = async (): Promise<void> => {\n\t\tconst startedMs = Date.now();\n\t\ttry {\n\t\t\tconst chunkHits = await service!.searchChunks(query, EMBED_TOP_K, glob);\n\t\t\tconst hits: RankedHit[] = chunkHits.map((hit, i) => ({\n\t\t\t\tid: hit.id,\n\t\t\t\trank: i + 1,\n\t\t\t\tscore: hit.score,\n\t\t\t\tsource: \"embed\",\n\t\t\t}));\n\t\t\tfor (const hit of chunkHits) {\n\t\t\t\tspans.set(hit.id, { path: hit.path, startLine: hit.startLine, endLine: hit.endLine });\n\t\t\t}\n\t\t\tlists.push(hits);\n\t\t\tretrieverStats.embed = { latencyMs: Date.now() - startedMs, hitCount: hits.length };\n\t\t} catch (e) {\n\t\t\terrors.push(e instanceof Error ? e : new Error(String(e)));\n\t\t\tretrieverStats.embed = { latencyMs: Date.now() - startedMs, hitCount: 0 };\n\t\t}\n\t};\n\n\tconst runs: Promise<void>[] = [];\n\tif (mode === \"lexical\" || mode === \"hybrid\") runs.push(runLexical());\n\tif (mode === \"semantic\" || mode === \"hybrid\") runs.push(runEmbed());\n\tawait Promise.all(runs);\n\tif (signal?.aborted) throw new Error(\"Operation aborted\");\n\t// A partial failure in hybrid degrades to whichever retriever survived;\n\t// only a total loss is an error.\n\tif (lists.length === 0) throw errors[0] ?? new Error(\"search produced no retriever results\");\n\n\tconst fused = rrfFuse(lists, rrfK).slice(0, FUSED_WINDOW);\n\tlet candidates: FusedCandidate[] = [];\n\tfor (const hit of fused) {\n\t\tconst span = spans.get(hit.id);\n\t\tif (span) candidates.push({ ...hit, ...span });\n\t}\n\n\tlet rerankInfo: SearchTrace[\"rerank\"];\n\tif (options.rerank !== false) {\n\t\tconst reranked = rerankCandidates(query, candidates, cwd);\n\t\trerankInfo = { applied: true, candidateCount: candidates.length, latencyMs: reranked.latencyMs };\n\t\tcandidates = reranked.candidates;\n\t}\n\tcandidates = candidates.slice(0, limit);\n\n\treturn {\n\t\tcandidates,\n\t\tresolvedMode: mode,\n\t\tdegradedReason: resolution.degradedReason,\n\t\tindexPhase: state?.phase === \"ready\" ? \"ready\" : state?.phase === \"indexing\" ? \"indexing\" : \"unavailable\",\n\t\tretrievers: retrieverStats,\n\t\tindexing: state?.phase === \"indexing\" ? { done: state.done, total: state.total } : undefined,\n\t\trrfK,\n\t\trerank: rerankInfo,\n\t};\n}\n\nexport async function runSearch(options: RunSearchOptions): Promise<RunSearchResult> {\n\tconst retrieved = await retrieveCandidates(options);\n\n\tconst assembled = assembleContext(retrieved.candidates, { cwd: options.cwd, tokenBudget: options.tokenBudget });\n\n\twriteSearchTrace(options.cwd, {\n\t\ttimestampMs: Date.now(),\n\t\tquery: options.query,\n\t\trequestedMode: options.mode ?? \"auto\",\n\t\tresolvedMode: retrieved.resolvedMode,\n\t\tdegradedReason: retrieved.degradedReason,\n\t\tindexPhase: retrieved.indexPhase,\n\t\trrfK: retrieved.resolvedMode === \"hybrid\" ? retrieved.rrfK : undefined,\n\t\tretrievers: retrieved.retrievers,\n\t\tfused: retrieved.candidates.map(({ id, rrfScore, ranks, rawScores }) => ({ id, rrfScore, ranks, rawScores })),\n\t\trerank: retrieved.rerank,\n\t});\n\n\treturn {\n\t\ttext: assembled.text,\n\t\tresolvedMode: retrieved.resolvedMode,\n\t\tdegradedReason: retrieved.degradedReason,\n\t\tresultCount: retrieved.candidates.length,\n\t\tindexing: retrieved.indexing,\n\t};\n}\n"]}
+{"version":3,"file":"hybrid-search.d.ts","sourceRoot":"","sources":["../../../src/core/search/hybrid-search.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,mCAAmC,CAAC;AAQ1E,OAAO,KAAK,EAAiB,cAAc,EAAa,kBAAkB,EAAE,UAAU,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAcxH,MAAM,WAAW,eAAe;IAC/B,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,UAAU,CAAC;IAClB,kDAAkD;IAClD,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,yCAAyC;IACzC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,wFAAwF;IACxF,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,wEAAwE;IACxE,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,OAAO,CAAC,EAAE,gBAAgB,CAAC;IAC3B,MAAM,CAAC,EAAE,WAAW,CAAC;CACrB;AAED,MAAM,WAAW,cAAc;IAC9B,UAAU,EAAE,cAAc,EAAE,CAAC;IAC7B,YAAY,EAAE,kBAAkB,CAAC;IACjC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,UAAU,EAAE,WAAW,CAAC,YAAY,CAAC,CAAC;IACtC,UAAU,EAAE,WAAW,CAAC,YAAY,CAAC,CAAC;IACtC,uDAAuD;IACvD,QAAQ,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAC3C,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,CAAC,EAAE,WAAW,CAAC,QAAQ,CAAC,CAAC;CAC/B;AAED,MAAM,WAAW,gBAAiB,SAAQ,eAAe;IACxD,oDAAoD;IACpD,WAAW,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,eAAe;IAC/B,IAAI,EAAE,MAAM,CAAC;IACb,YAAY,EAAE,kBAAkB,CAAC;IACjC,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,WAAW,EAAE,MAAM,CAAC;IACpB,uDAAuD;IACvD,QAAQ,CAAC,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;CAC3C;AAYD,wBAAsB,kBAAkB,CAAC,OAAO,EAAE,eAAe,GAAG,OAAO,CAAC,cAAc,CAAC,CA2G1F;AAED,wBAAsB,SAAS,CAAC,OAAO,EAAE,gBAAgB,GAAG,OAAO,CAAC,eAAe,CAAC,CAyBnF","sourcesContent":["/**\n * Hybrid search orchestrator: resolve mode, run retrievers in parallel, fuse\n * by rank, expand within budget, trace everything\n * (docs/hybrid-retrieval-design.md).\n *\n * Single-retriever modes flow through the same pipeline — rrfFuse over one\n * list preserves its order — so lexical, semantic, and hybrid all produce the\n * same result shape and the same trace record.\n *\n * `retrieveCandidates` is the candidate-level core (also used by the eval\n * harness, which needs forced modes, a configurable `k`, and no trace\n * pollution); `runSearch` wraps it with span expansion and tracing for the\n * tool.\n */\n\nimport type { EmbsearchService } from \"../embsearch/embsearch-service.js\";\nimport { adaptGrepHits, type ChunkLookup } from \"./adapter.js\";\nimport { assembleContext } from \"./context-assembler.js\";\nimport { runLexicalRetriever } from \"./lexical-retriever.js\";\nimport { resolveSearchMode } from \"./mode.js\";\nimport { rerankCandidates } from \"./rerank.js\";\nimport { DEFAULT_RRF_K, rrfFuse } from \"./rrf.js\";\nimport { writeSearchTrace } from \"./trace.js\";\nimport type { CandidateSpan, FusedCandidate, RankedHit, ResolvedSearchMode, SearchMode, SearchTrace } from \"./types.js\";\n\n/** Raw grep line-hits fetched per query (pre-collapse). */\nconst LEXICAL_MATCH_LIMIT = 200;\n/** Adapted lexical candidates entering fusion. The eval gate showed the\n *  uncapped lexical tail diluting hybrid below plain semantic: lexical\n *  precision is front-loaded by the adapter's term-evidence ranking, while\n *  RRF weighs a rank-30 lexical candidate like a rank-30 embedding hit. */\nconst LEXICAL_FUSION_CAP = 20;\n/** Embedding hits fetched per query — deep enough for fusion to matter. */\nconst EMBED_TOP_K = 50;\n/** Fused candidates kept for reranking / final slicing. */\nconst FUSED_WINDOW = 50;\n\nexport interface RetrieveOptions {\n\tcwd: string;\n\tquery: string;\n\tmode?: SearchMode;\n\t/** Optional glob filter applied to file paths. */\n\tglob?: string;\n\t/** Maximum fused candidates returned. */\n\tlimit?: number;\n\t/** RRF constant override (eval harness sweeps this). Default: {@link DEFAULT_RRF_K}. */\n\trrfK?: number;\n\t/** Rerank the fused top-50 before slicing to `limit`. Default: true. */\n\trerank?: boolean;\n\tservice?: EmbsearchService;\n\tsignal?: AbortSignal;\n}\n\nexport interface RetrieveResult {\n\tcandidates: FusedCandidate[];\n\tresolvedMode: ResolvedSearchMode;\n\tdegradedReason?: string;\n\tindexPhase: SearchTrace[\"indexPhase\"];\n\tretrievers: SearchTrace[\"retrievers\"];\n\t/** Set while the embedding index is still building. */\n\tindexing?: { done: number; total: number };\n\trrfK: number;\n\trerank?: SearchTrace[\"rerank\"];\n}\n\nexport interface RunSearchOptions extends RetrieveOptions {\n\t/** Approximate token budget for the result text. */\n\ttokenBudget?: number;\n}\n\nexport interface RunSearchResult {\n\ttext: string;\n\tresolvedMode: ResolvedSearchMode;\n\tdegradedReason?: string;\n\tresultCount: number;\n\t/** Set while the embedding index is still building. */\n\tindexing?: { done: number; total: number };\n}\n\nfunction normalizeSearchGlob(glob: string | undefined): string | undefined {\n\tif (!glob) return undefined;\n\t// Match fd/rg semantics: a slash-containing glob is anchored anywhere in\n\t// the tree, so prepend \"**/\" unless it already starts with a slash or \"**/\".\n\tif (glob.includes(\"/\") && !glob.startsWith(\"/\") && !glob.startsWith(\"**/\")) {\n\t\treturn `**/${glob}`;\n\t}\n\treturn glob;\n}\n\nexport async function retrieveCandidates(options: RetrieveOptions): Promise<RetrieveResult> {\n\tconst { cwd, query, service, signal } = options;\n\tconst glob = normalizeSearchGlob(options.glob);\n\tconst requestedMode = options.mode ?? \"auto\";\n\tconst limit = Math.max(1, options.limit ?? 10);\n\tconst rrfK = options.rrfK ?? DEFAULT_RRF_K;\n\n\tconst state = service?.getState();\n\tconst embedAvailable = service?.isAvailable() ?? false;\n\tconst embedUnavailableReason =\n\t\tstate === undefined\n\t\t\t? \"semantic index is not enabled\"\n\t\t\t: state.phase === \"unavailable\" || state.phase === \"skipped\"\n\t\t\t\t? state.reason\n\t\t\t\t: state.phase === \"idle\"\n\t\t\t\t\t? \"semantic index has not started\"\n\t\t\t\t\t: undefined;\n\n\tconst resolution = resolveSearchMode(query, requestedMode, embedAvailable, embedUnavailableReason);\n\tconst mode = resolution.mode;\n\n\t// Map lexical hits onto indexed chunk ids whenever the sidecar is usable,\n\t// even in lexical-only mode, so identities line up across modes.\n\tconst lookupChunk: ChunkLookup | undefined = embedAvailable\n\t\t? (rel, line) => service!.findEnclosingChunk(rel, line)\n\t\t: undefined;\n\n\tconst spans = new Map<string, CandidateSpan>();\n\tconst lists: RankedHit[][] = [];\n\tconst retrieverStats: SearchTrace[\"retrievers\"] = {};\n\tconst errors: Error[] = [];\n\n\tconst runLexical = async (): Promise<void> => {\n\t\tconst startedMs = Date.now();\n\t\ttry {\n\t\t\tconst lineHits = await runLexicalRetriever({ cwd, query, limit: LEXICAL_MATCH_LIMIT, glob, signal });\n\t\t\tconst adapted = adaptGrepHits(lineHits, lookupChunk);\n\t\t\t// In single-retriever lexical mode the full list is the result; in\n\t\t\t// hybrid, only the front-loaded head is trustworthy enough to vote.\n\t\t\tconst hits = mode === \"hybrid\" ? adapted.hits.slice(0, LEXICAL_FUSION_CAP) : adapted.hits;\n\t\t\tfor (const [id, span] of adapted.spans) if (!spans.has(id)) spans.set(id, span);\n\t\t\tlists.push(hits);\n\t\t\tretrieverStats.grep = { latencyMs: Date.now() - startedMs, hitCount: hits.length };\n\t\t} catch (e) {\n\t\t\terrors.push(e instanceof Error ? e : new Error(String(e)));\n\t\t\tretrieverStats.grep = { latencyMs: Date.now() - startedMs, hitCount: 0 };\n\t\t}\n\t};\n\n\tconst runEmbed = async (): Promise<void> => {\n\t\tconst startedMs = Date.now();\n\t\ttry {\n\t\t\t// The flat index pads top-k with whatever exists; with the cosine\n\t\t\t// metric the store uses, score <= 0 means \"no relation at all\", so\n\t\t\t// those padding hits would cast RRF votes on pure noise.\n\t\t\tconst chunkHits = (await service!.searchChunks(query, EMBED_TOP_K, glob)).filter((hit) => hit.score > 0);\n\t\t\tconst hits: RankedHit[] = chunkHits.map((hit, i) => ({\n\t\t\t\tid: hit.id,\n\t\t\t\trank: i + 1,\n\t\t\t\tscore: hit.score,\n\t\t\t\tsource: \"embed\",\n\t\t\t}));\n\t\t\tfor (const hit of chunkHits) {\n\t\t\t\tspans.set(hit.id, { path: hit.path, startLine: hit.startLine, endLine: hit.endLine });\n\t\t\t}\n\t\t\tlists.push(hits);\n\t\t\tretrieverStats.embed = { latencyMs: Date.now() - startedMs, hitCount: hits.length };\n\t\t} catch (e) {\n\t\t\terrors.push(e instanceof Error ? e : new Error(String(e)));\n\t\t\tretrieverStats.embed = { latencyMs: Date.now() - startedMs, hitCount: 0 };\n\t\t}\n\t};\n\n\tconst runs: Promise<void>[] = [];\n\tif (mode === \"lexical\" || mode === \"hybrid\") runs.push(runLexical());\n\tif (mode === \"semantic\" || mode === \"hybrid\") runs.push(runEmbed());\n\tawait Promise.all(runs);\n\tif (signal?.aborted) throw new Error(\"Operation aborted\");\n\t// A partial failure in hybrid degrades to whichever retriever survived;\n\t// only a total loss is an error.\n\tif (lists.length === 0) throw errors[0] ?? new Error(\"search produced no retriever results\");\n\n\tconst fused = rrfFuse(lists, rrfK).slice(0, FUSED_WINDOW);\n\tlet candidates: FusedCandidate[] = [];\n\tfor (const hit of fused) {\n\t\tconst span = spans.get(hit.id);\n\t\tif (span) candidates.push({ ...hit, ...span });\n\t}\n\n\tlet rerankInfo: SearchTrace[\"rerank\"];\n\tif (options.rerank !== false) {\n\t\tconst reranked = rerankCandidates(query, candidates, cwd);\n\t\trerankInfo = { applied: true, candidateCount: candidates.length, latencyMs: reranked.latencyMs };\n\t\tcandidates = reranked.candidates;\n\t}\n\tcandidates = candidates.slice(0, limit);\n\n\treturn {\n\t\tcandidates,\n\t\tresolvedMode: mode,\n\t\tdegradedReason: resolution.degradedReason,\n\t\tindexPhase: state?.phase === \"ready\" ? \"ready\" : state?.phase === \"indexing\" ? \"indexing\" : \"unavailable\",\n\t\tretrievers: retrieverStats,\n\t\tindexing: state?.phase === \"indexing\" ? { done: state.done, total: state.total } : undefined,\n\t\trrfK,\n\t\trerank: rerankInfo,\n\t};\n}\n\nexport async function runSearch(options: RunSearchOptions): Promise<RunSearchResult> {\n\tconst retrieved = await retrieveCandidates(options);\n\n\tconst assembled = assembleContext(retrieved.candidates, { cwd: options.cwd, tokenBudget: options.tokenBudget });\n\n\twriteSearchTrace(options.cwd, {\n\t\ttimestampMs: Date.now(),\n\t\tquery: options.query,\n\t\trequestedMode: options.mode ?? \"auto\",\n\t\tresolvedMode: retrieved.resolvedMode,\n\t\tdegradedReason: retrieved.degradedReason,\n\t\tindexPhase: retrieved.indexPhase,\n\t\trrfK: retrieved.resolvedMode === \"hybrid\" ? retrieved.rrfK : undefined,\n\t\tretrievers: retrieved.retrievers,\n\t\tfused: retrieved.candidates.map(({ id, rrfScore, ranks, rawScores }) => ({ id, rrfScore, ranks, rawScores })),\n\t\trerank: retrieved.rerank,\n\t});\n\n\treturn {\n\t\ttext: assembled.text,\n\t\tresolvedMode: retrieved.resolvedMode,\n\t\tdegradedReason: retrieved.degradedReason,\n\t\tresultCount: retrieved.candidates.length,\n\t\tindexing: retrieved.indexing,\n\t};\n}\n"]}

package/dist/core/search/hybrid-search.js

@@ -88,7 +88,10 @@ export async function retrieveCandidates(options) {
     const runEmbed = async () => {
         const startedMs = Date.now();
         try {
-            const chunkHits = await service.searchChunks(query, EMBED_TOP_K, glob);
+            // The flat index pads top-k with whatever exists; with the cosine
+            // metric the store uses, score <= 0 means "no relation at all", so
+            // those padding hits would cast RRF votes on pure noise.
+            const chunkHits = (await service.searchChunks(query, EMBED_TOP_K, glob)).filter((hit) => hit.score > 0);
             const hits = chunkHits.map((hit, i) => ({
                 id: hit.id,
                 rank: i + 1,

package/dist/core/search/hybrid-search.js.map

@@ -1 +1 @@
-{"version":3,"file":"hybrid-search.js","sourceRoot":"","sources":["../../../src/core/search/hybrid-search.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAGH,OAAO,EAAE,aAAa,EAAoB,MAAM,cAAc,CAAC;AAC/D,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAC7D,OAAO,EAAE,iBAAiB,EAAE,MAAM,WAAW,CAAC;AAC9C,OAAO,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAC/C,OAAO,EAAE,aAAa,EAAE,OAAO,EAAE,MAAM,UAAU,CAAC;AAClD,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAG9C,2DAA2D;AAC3D,MAAM,mBAAmB,GAAG,GAAG,CAAC;AAChC;;;2EAG2E;AAC3E,MAAM,kBAAkB,GAAG,EAAE,CAAC;AAC9B,6EAA2E;AAC3E,MAAM,WAAW,GAAG,EAAE,CAAC;AACvB,2DAA2D;AAC3D,MAAM,YAAY,GAAG,EAAE,CAAC;AA4CxB,SAAS,mBAAmB,CAAC,IAAwB,EAAsB;IAC1E,IAAI,CAAC,IAAI;QAAE,OAAO,SAAS,CAAC;IAC5B,yEAAyE;IACzE,6EAA6E;IAC7E,IAAI,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;QAC5E,OAAO,MAAM,IAAI,EAAE,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,OAAwB,EAA2B;IAC3F,MAAM,EAAE,GAAG,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC;IAChD,MAAM,IAAI,GAAG,mBAAmB,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IAC/C,MAAM,aAAa,GAAG,OAAO,CAAC,IAAI,IAAI,MAAM,CAAC;IAC7C,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC;IAC/C,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,IAAI,aAAa,CAAC;IAE3C,MAAM,KAAK,GAAG,OAAO,EAAE,QAAQ,EAAE,CAAC;IAClC,MAAM,cAAc,GAAG,OAAO,EAAE,WAAW,EAAE,IAAI,KAAK,CAAC;IACvD,MAAM,sBAAsB,GAC3B,KAAK,KAAK,SAAS;QAClB,CAAC,CAAC,+BAA+B;QACjC,CAAC,CAAC,KAAK,CAAC,KAAK,KAAK,aAAa,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS;YAC3D,CAAC,CAAC,KAAK,CAAC,MAAM;YACd,CAAC,CAAC,KAAK,CAAC,KAAK,KAAK,MAAM;gBACvB,CAAC,CAAC,gCAAgC;gBAClC,CAAC,CAAC,SAAS,CAAC;IAEhB,MAAM,UAAU,GAAG,iBAAiB,CAAC,KAAK,EAAE,aAAa,EAAE,cAAc,EAAE,sBAAsB,CAAC,CAAC;IACnG,MAAM,IAAI,GAAG,UAAU,CAAC,IAAI,CAAC;IAE7B,0EAA0E;IAC1E,iEAAiE;IACjE,MAAM,WAAW,GAA4B,cAAc;QAC1D,CAAC,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,OAAQ,CAAC,kBAAkB,CAAC,GAAG,EAAE,IAAI,CAAC;QACvD,CAAC,CAAC,SAAS,CAAC;IAEb,MAAM,KAAK,GAAG,IAAI,GAAG,EAAyB,CAAC;IAC/C,MAAM,KAAK,GAAkB,EAAE,CAAC;IAChC,MAAM,cAAc,GAA8B,EAAE,CAAC;IACrD,MAAM,MAAM,GAAY,EAAE,CAAC;IAE3B,MAAM,UAAU,GAAG,KAAK,IAAmB,EAAE,CAAC;QAC7C,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,IAAI,CAAC;YACJ,MAAM,QAAQ,GAAG,MAAM,mBAAmB,CAAC,EAAE,GAAG,EAAE,KAAK,EAAE,KAAK,EAAE,mBAAmB,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;YACrG,MAAM,OAAO,GAAG,aAAa,CAAC,QAAQ,EAAE,WAAW,CAAC,CAAC;YACrD,mEAAmE;YACnE,oEAAoE;YACpE,MAAM,IAAI,GAAG,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,kBAAkB,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC;YAC1F,KAAK,MAAM,CAAC,EAAE,EAAE,IAAI,CAAC,IAAI,OAAO,CAAC,KAAK;gBAAE,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;oBAAE,KAAK,CAAC,GAAG,CAAC,EAAE,EAAE,IAAI,CAAC,CAAC;YAChF,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACjB,cAAc,CAAC,IAAI,GAAG,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,QAAQ,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC;QACpF,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACZ,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC3D,cAAc,CAAC,IAAI,GAAG,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;QAC1E,CAAC;IAAA,CACD,CAAC;IAEF,MAAM,QAAQ,GAAG,KAAK,IAAmB,EAAE,CAAC;QAC3C,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,IAAI,CAAC;YACJ,MAAM,SAAS,GAAG,MAAM,OAAQ,CAAC,YAAY,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,CAAC,CAAC;YACxE,MAAM,IAAI,GAAgB,SAAS,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;gBACpD,EAAE,EAAE,GAAG,CAAC,EAAE;gBACV,IAAI,EAAE,CAAC,GAAG,CAAC;gBACX,KAAK,EAAE,GAAG,CAAC,KAAK;gBAChB,MAAM,EAAE,OAAO;aACf,CAAC,CAAC,CAAC;YACJ,KAAK,MAAM,GAAG,IAAI,SAAS,EAAE,CAAC;gBAC7B,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,SAAS,EAAE,GAAG,CAAC,SAAS,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC;YACvF,CAAC;YACD,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACjB,cAAc,CAAC,KAAK,GAAG,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,QAAQ,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC;QACrF,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACZ,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC3D,cAAc,CAAC,KAAK,GAAG,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;QAC3E,CAAC;IAAA,CACD,CAAC;IAEF,MAAM,IAAI,GAAoB,EAAE,CAAC;IACjC,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,KAAK,QAAQ;QAAE,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC;IACrE,IAAI,IAAI,KAAK,UAAU,IAAI,IAAI,KAAK,QAAQ;QAAE,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC;IACpE,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACxB,IAAI,MAAM,EAAE,OAAO;QAAE,MAAM,IAAI,KAAK,CAAC,mBAAmB,CAAC,CAAC;IAC1D,wEAAwE;IACxE,iCAAiC;IACjC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,MAAM,MAAM,CAAC,CAAC,CAAC,IAAI,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAC;IAE7F,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,YAAY,CAAC,CAAC;IAC1D,IAAI,UAAU,GAAqB,EAAE,CAAC;IACtC,KAAK,MAAM,GAAG,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,IAAI,GAAG,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAC/B,IAAI,IAAI;YAAE,UAAU,CAAC,IAAI,CAAC,EAAE,GAAG,GAAG,EAAE,GAAG,IAAI,EAAE,CAAC,CAAC;IAChD,CAAC;IAED,IAAI,UAAiC,CAAC;IACtC,IAAI,OAAO,CAAC,MAAM,KAAK,KAAK,EAAE,CAAC;QAC9B,MAAM,QAAQ,GAAG,gBAAgB,CAAC,KAAK,EAAE,UAAU,EAAE,GAAG,CAAC,CAAC;QAC1D,UAAU,GAAG,EAAE,OAAO,EAAE,IAAI,EAAE,cAAc,EAAE,UAAU,CAAC,MAAM,EAAE,SAAS,EAAE,QAAQ,CAAC,SAAS,EAAE,CAAC;QACjG,UAAU,GAAG,QAAQ,CAAC,UAAU,CAAC;IAClC,CAAC;IACD,UAAU,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;IAExC,OAAO;QACN,UAAU;QACV,YAAY,EAAE,IAAI;QAClB,cAAc,EAAE,UAAU,CAAC,cAAc;QACzC,UAAU,EAAE,KAAK,EAAE,KAAK,KAAK,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,EAAE,KAAK,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,aAAa;QACzG,UAAU,EAAE,cAAc;QAC1B,QAAQ,EAAE,KAAK,EAAE,KAAK,KAAK,UAAU,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,SAAS;QAC5F,IAAI;QACJ,MAAM,EAAE,UAAU;KAClB,CAAC;AAAA,CACF;AAED,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,OAAyB,EAA4B;IACpF,MAAM,SAAS,GAAG,MAAM,kBAAkB,CAAC,OAAO,CAAC,CAAC;IAEpD,MAAM,SAAS,GAAG,eAAe,CAAC,SAAS,CAAC,UAAU,EAAE,EAAE,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE,WAAW,EAAE,OAAO,CAAC,WAAW,EAAE,CAAC,CAAC;IAEhH,gBAAgB,CAAC,OAAO,CAAC,GAAG,EAAE;QAC7B,WAAW,EAAE,IAAI,CAAC,GAAG,EAAE;QACvB,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,aAAa,EAAE,OAAO,CAAC,IAAI,IAAI,MAAM;QACrC,YAAY,EAAE,SAAS,CAAC,YAAY;QACpC,cAAc,EAAE,SAAS,CAAC,cAAc;QACxC,UAAU,EAAE,SAAS,CAAC,UAAU;QAChC,IAAI,EAAE,SAAS,CAAC,YAAY,KAAK,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS;QACtE,UAAU,EAAE,SAAS,CAAC,UAAU;QAChC,KAAK,EAAE,SAAS,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,CAAC,CAAC;QAC7G,MAAM,EAAE,SAAS,CAAC,MAAM;KACxB,CAAC,CAAC;IAEH,OAAO;QACN,IAAI,EAAE,SAAS,CAAC,IAAI;QACpB,YAAY,EAAE,SAAS,CAAC,YAAY;QACpC,cAAc,EAAE,SAAS,CAAC,cAAc;QACxC,WAAW,EAAE,SAAS,CAAC,UAAU,CAAC,MAAM;QACxC,QAAQ,EAAE,SAAS,CAAC,QAAQ;KAC5B,CAAC;AAAA,CACF","sourcesContent":["/**\n * Hybrid search orchestrator: resolve mode, run retrievers in parallel, fuse\n * by rank, expand within budget, trace everything\n * (docs/hybrid-retrieval-design.md).\n *\n * Single-retriever modes flow through the same pipeline — rrfFuse over one\n * list preserves its order — so lexical, semantic, and hybrid all produce the\n * same result shape and the same trace record.\n *\n * `retrieveCandidates` is the candidate-level core (also used by the eval\n * harness, which needs forced modes, a configurable `k`, and no trace\n * pollution); `runSearch` wraps it with span expansion and tracing for the\n * tool.\n */\n\nimport type { EmbsearchService } from \"../embsearch/embsearch-service.js\";\nimport { adaptGrepHits, type ChunkLookup } from \"./adapter.js\";\nimport { assembleContext } from \"./context-assembler.js\";\nimport { runLexicalRetriever } from \"./lexical-retriever.js\";\nimport { resolveSearchMode } from \"./mode.js\";\nimport { rerankCandidates } from \"./rerank.js\";\nimport { DEFAULT_RRF_K, rrfFuse } from \"./rrf.js\";\nimport { writeSearchTrace } from \"./trace.js\";\nimport type { CandidateSpan, FusedCandidate, RankedHit, ResolvedSearchMode, SearchMode, SearchTrace } from \"./types.js\";\n\n/** Raw grep line-hits fetched per query (pre-collapse). */\nconst LEXICAL_MATCH_LIMIT = 200;\n/** Adapted lexical candidates entering fusion. The eval gate showed the\n *  uncapped lexical tail diluting hybrid below plain semantic: lexical\n *  precision is front-loaded by the adapter's term-evidence ranking, while\n *  RRF weighs a rank-30 lexical candidate like a rank-30 embedding hit. */\nconst LEXICAL_FUSION_CAP = 20;\n/** Embedding hits fetched per query — deep enough for fusion to matter. */\nconst EMBED_TOP_K = 50;\n/** Fused candidates kept for reranking / final slicing. */\nconst FUSED_WINDOW = 50;\n\nexport interface RetrieveOptions {\n\tcwd: string;\n\tquery: string;\n\tmode?: SearchMode;\n\t/** Optional glob filter applied to file paths. */\n\tglob?: string;\n\t/** Maximum fused candidates returned. */\n\tlimit?: number;\n\t/** RRF constant override (eval harness sweeps this). Default: {@link DEFAULT_RRF_K}. */\n\trrfK?: number;\n\t/** Rerank the fused top-50 before slicing to `limit`. Default: true. */\n\trerank?: boolean;\n\tservice?: EmbsearchService;\n\tsignal?: AbortSignal;\n}\n\nexport interface RetrieveResult {\n\tcandidates: FusedCandidate[];\n\tresolvedMode: ResolvedSearchMode;\n\tdegradedReason?: string;\n\tindexPhase: SearchTrace[\"indexPhase\"];\n\tretrievers: SearchTrace[\"retrievers\"];\n\t/** Set while the embedding index is still building. */\n\tindexing?: { done: number; total: number };\n\trrfK: number;\n\trerank?: SearchTrace[\"rerank\"];\n}\n\nexport interface RunSearchOptions extends RetrieveOptions {\n\t/** Approximate token budget for the result text. */\n\ttokenBudget?: number;\n}\n\nexport interface RunSearchResult {\n\ttext: string;\n\tresolvedMode: ResolvedSearchMode;\n\tdegradedReason?: string;\n\tresultCount: number;\n\t/** Set while the embedding index is still building. */\n\tindexing?: { done: number; total: number };\n}\n\nfunction normalizeSearchGlob(glob: string | undefined): string | undefined {\n\tif (!glob) return undefined;\n\t// Match fd/rg semantics: a slash-containing glob is anchored anywhere in\n\t// the tree, so prepend \"**/\" unless it already starts with a slash or \"**/\".\n\tif (glob.includes(\"/\") && !glob.startsWith(\"/\") && !glob.startsWith(\"**/\")) {\n\t\treturn `**/${glob}`;\n\t}\n\treturn glob;\n}\n\nexport async function retrieveCandidates(options: RetrieveOptions): Promise<RetrieveResult> {\n\tconst { cwd, query, service, signal } = options;\n\tconst glob = normalizeSearchGlob(options.glob);\n\tconst requestedMode = options.mode ?? \"auto\";\n\tconst limit = Math.max(1, options.limit ?? 10);\n\tconst rrfK = options.rrfK ?? DEFAULT_RRF_K;\n\n\tconst state = service?.getState();\n\tconst embedAvailable = service?.isAvailable() ?? false;\n\tconst embedUnavailableReason =\n\t\tstate === undefined\n\t\t\t? \"semantic index is not enabled\"\n\t\t\t: state.phase === \"unavailable\" || state.phase === \"skipped\"\n\t\t\t\t? state.reason\n\t\t\t\t: state.phase === \"idle\"\n\t\t\t\t\t? \"semantic index has not started\"\n\t\t\t\t\t: undefined;\n\n\tconst resolution = resolveSearchMode(query, requestedMode, embedAvailable, embedUnavailableReason);\n\tconst mode = resolution.mode;\n\n\t// Map lexical hits onto indexed chunk ids whenever the sidecar is usable,\n\t// even in lexical-only mode, so identities line up across modes.\n\tconst lookupChunk: ChunkLookup | undefined = embedAvailable\n\t\t? (rel, line) => service!.findEnclosingChunk(rel, line)\n\t\t: undefined;\n\n\tconst spans = new Map<string, CandidateSpan>();\n\tconst lists: RankedHit[][] = [];\n\tconst retrieverStats: SearchTrace[\"retrievers\"] = {};\n\tconst errors: Error[] = [];\n\n\tconst runLexical = async (): Promise<void> => {\n\t\tconst startedMs = Date.now();\n\t\ttry {\n\t\t\tconst lineHits = await runLexicalRetriever({ cwd, query, limit: LEXICAL_MATCH_LIMIT, glob, signal });\n\t\t\tconst adapted = adaptGrepHits(lineHits, lookupChunk);\n\t\t\t// In single-retriever lexical mode the full list is the result; in\n\t\t\t// hybrid, only the front-loaded head is trustworthy enough to vote.\n\t\t\tconst hits = mode === \"hybrid\" ? adapted.hits.slice(0, LEXICAL_FUSION_CAP) : adapted.hits;\n\t\t\tfor (const [id, span] of adapted.spans) if (!spans.has(id)) spans.set(id, span);\n\t\t\tlists.push(hits);\n\t\t\tretrieverStats.grep = { latencyMs: Date.now() - startedMs, hitCount: hits.length };\n\t\t} catch (e) {\n\t\t\terrors.push(e instanceof Error ? e : new Error(String(e)));\n\t\t\tretrieverStats.grep = { latencyMs: Date.now() - startedMs, hitCount: 0 };\n\t\t}\n\t};\n\n\tconst runEmbed = async (): Promise<void> => {\n\t\tconst startedMs = Date.now();\n\t\ttry {\n\t\t\tconst chunkHits = await service!.searchChunks(query, EMBED_TOP_K, glob);\n\t\t\tconst hits: RankedHit[] = chunkHits.map((hit, i) => ({\n\t\t\t\tid: hit.id,\n\t\t\t\trank: i + 1,\n\t\t\t\tscore: hit.score,\n\t\t\t\tsource: \"embed\",\n\t\t\t}));\n\t\t\tfor (const hit of chunkHits) {\n\t\t\t\tspans.set(hit.id, { path: hit.path, startLine: hit.startLine, endLine: hit.endLine });\n\t\t\t}\n\t\t\tlists.push(hits);\n\t\t\tretrieverStats.embed = { latencyMs: Date.now() - startedMs, hitCount: hits.length };\n\t\t} catch (e) {\n\t\t\terrors.push(e instanceof Error ? e : new Error(String(e)));\n\t\t\tretrieverStats.embed = { latencyMs: Date.now() - startedMs, hitCount: 0 };\n\t\t}\n\t};\n\n\tconst runs: Promise<void>[] = [];\n\tif (mode === \"lexical\" || mode === \"hybrid\") runs.push(runLexical());\n\tif (mode === \"semantic\" || mode === \"hybrid\") runs.push(runEmbed());\n\tawait Promise.all(runs);\n\tif (signal?.aborted) throw new Error(\"Operation aborted\");\n\t// A partial failure in hybrid degrades to whichever retriever survived;\n\t// only a total loss is an error.\n\tif (lists.length === 0) throw errors[0] ?? new Error(\"search produced no retriever results\");\n\n\tconst fused = rrfFuse(lists, rrfK).slice(0, FUSED_WINDOW);\n\tlet candidates: FusedCandidate[] = [];\n\tfor (const hit of fused) {\n\t\tconst span = spans.get(hit.id);\n\t\tif (span) candidates.push({ ...hit, ...span });\n\t}\n\n\tlet rerankInfo: SearchTrace[\"rerank\"];\n\tif (options.rerank !== false) {\n\t\tconst reranked = rerankCandidates(query, candidates, cwd);\n\t\trerankInfo = { applied: true, candidateCount: candidates.length, latencyMs: reranked.latencyMs };\n\t\tcandidates = reranked.candidates;\n\t}\n\tcandidates = candidates.slice(0, limit);\n\n\treturn {\n\t\tcandidates,\n\t\tresolvedMode: mode,\n\t\tdegradedReason: resolution.degradedReason,\n\t\tindexPhase: state?.phase === \"ready\" ? \"ready\" : state?.phase === \"indexing\" ? \"indexing\" : \"unavailable\",\n\t\tretrievers: retrieverStats,\n\t\tindexing: state?.phase === \"indexing\" ? { done: state.done, total: state.total } : undefined,\n\t\trrfK,\n\t\trerank: rerankInfo,\n\t};\n}\n\nexport async function runSearch(options: RunSearchOptions): Promise<RunSearchResult> {\n\tconst retrieved = await retrieveCandidates(options);\n\n\tconst assembled = assembleContext(retrieved.candidates, { cwd: options.cwd, tokenBudget: options.tokenBudget });\n\n\twriteSearchTrace(options.cwd, {\n\t\ttimestampMs: Date.now(),\n\t\tquery: options.query,\n\t\trequestedMode: options.mode ?? \"auto\",\n\t\tresolvedMode: retrieved.resolvedMode,\n\t\tdegradedReason: retrieved.degradedReason,\n\t\tindexPhase: retrieved.indexPhase,\n\t\trrfK: retrieved.resolvedMode === \"hybrid\" ? retrieved.rrfK : undefined,\n\t\tretrievers: retrieved.retrievers,\n\t\tfused: retrieved.candidates.map(({ id, rrfScore, ranks, rawScores }) => ({ id, rrfScore, ranks, rawScores })),\n\t\trerank: retrieved.rerank,\n\t});\n\n\treturn {\n\t\ttext: assembled.text,\n\t\tresolvedMode: retrieved.resolvedMode,\n\t\tdegradedReason: retrieved.degradedReason,\n\t\tresultCount: retrieved.candidates.length,\n\t\tindexing: retrieved.indexing,\n\t};\n}\n"]}
+{"version":3,"file":"hybrid-search.js","sourceRoot":"","sources":["../../../src/core/search/hybrid-search.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAGH,OAAO,EAAE,aAAa,EAAoB,MAAM,cAAc,CAAC;AAC/D,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AACzD,OAAO,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAC;AAC7D,OAAO,EAAE,iBAAiB,EAAE,MAAM,WAAW,CAAC;AAC9C,OAAO,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAC/C,OAAO,EAAE,aAAa,EAAE,OAAO,EAAE,MAAM,UAAU,CAAC;AAClD,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAG9C,2DAA2D;AAC3D,MAAM,mBAAmB,GAAG,GAAG,CAAC;AAChC;;;2EAG2E;AAC3E,MAAM,kBAAkB,GAAG,EAAE,CAAC;AAC9B,6EAA2E;AAC3E,MAAM,WAAW,GAAG,EAAE,CAAC;AACvB,2DAA2D;AAC3D,MAAM,YAAY,GAAG,EAAE,CAAC;AA4CxB,SAAS,mBAAmB,CAAC,IAAwB,EAAsB;IAC1E,IAAI,CAAC,IAAI;QAAE,OAAO,SAAS,CAAC;IAC5B,yEAAyE;IACzE,6EAA6E;IAC7E,IAAI,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;QAC5E,OAAO,MAAM,IAAI,EAAE,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,OAAwB,EAA2B;IAC3F,MAAM,EAAE,GAAG,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC;IAChD,MAAM,IAAI,GAAG,mBAAmB,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;IAC/C,MAAM,aAAa,GAAG,OAAO,CAAC,IAAI,IAAI,MAAM,CAAC;IAC7C,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC;IAC/C,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,IAAI,aAAa,CAAC;IAE3C,MAAM,KAAK,GAAG,OAAO,EAAE,QAAQ,EAAE,CAAC;IAClC,MAAM,cAAc,GAAG,OAAO,EAAE,WAAW,EAAE,IAAI,KAAK,CAAC;IACvD,MAAM,sBAAsB,GAC3B,KAAK,KAAK,SAAS;QAClB,CAAC,CAAC,+BAA+B;QACjC,CAAC,CAAC,KAAK,CAAC,KAAK,KAAK,aAAa,IAAI,KAAK,CAAC,KAAK,KAAK,SAAS;YAC3D,CAAC,CAAC,KAAK,CAAC,MAAM;YACd,CAAC,CAAC,KAAK,CAAC,KAAK,KAAK,MAAM;gBACvB,CAAC,CAAC,gCAAgC;gBAClC,CAAC,CAAC,SAAS,CAAC;IAEhB,MAAM,UAAU,GAAG,iBAAiB,CAAC,KAAK,EAAE,aAAa,EAAE,cAAc,EAAE,sBAAsB,CAAC,CAAC;IACnG,MAAM,IAAI,GAAG,UAAU,CAAC,IAAI,CAAC;IAE7B,0EAA0E;IAC1E,iEAAiE;IACjE,MAAM,WAAW,GAA4B,cAAc;QAC1D,CAAC,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,OAAQ,CAAC,kBAAkB,CAAC,GAAG,EAAE,IAAI,CAAC;QACvD,CAAC,CAAC,SAAS,CAAC;IAEb,MAAM,KAAK,GAAG,IAAI,GAAG,EAAyB,CAAC;IAC/C,MAAM,KAAK,GAAkB,EAAE,CAAC;IAChC,MAAM,cAAc,GAA8B,EAAE,CAAC;IACrD,MAAM,MAAM,GAAY,EAAE,CAAC;IAE3B,MAAM,UAAU,GAAG,KAAK,IAAmB,EAAE,CAAC;QAC7C,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,IAAI,CAAC;YACJ,MAAM,QAAQ,GAAG,MAAM,mBAAmB,CAAC,EAAE,GAAG,EAAE,KAAK,EAAE,KAAK,EAAE,mBAAmB,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;YACrG,MAAM,OAAO,GAAG,aAAa,CAAC,QAAQ,EAAE,WAAW,CAAC,CAAC;YACrD,mEAAmE;YACnE,oEAAoE;YACpE,MAAM,IAAI,GAAG,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,kBAAkB,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC;YAC1F,KAAK,MAAM,CAAC,EAAE,EAAE,IAAI,CAAC,IAAI,OAAO,CAAC,KAAK;gBAAE,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;oBAAE,KAAK,CAAC,GAAG,CAAC,EAAE,EAAE,IAAI,CAAC,CAAC;YAChF,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACjB,cAAc,CAAC,IAAI,GAAG,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,QAAQ,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC;QACpF,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACZ,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC3D,cAAc,CAAC,IAAI,GAAG,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;QAC1E,CAAC;IAAA,CACD,CAAC;IAEF,MAAM,QAAQ,GAAG,KAAK,IAAmB,EAAE,CAAC;QAC3C,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,IAAI,CAAC;YACJ,kEAAkE;YAClE,mEAAmE;YACnE,yDAAyD;YACzD,MAAM,SAAS,GAAG,CAAC,MAAM,OAAQ,CAAC,YAAY,CAAC,KAAK,EAAE,WAAW,EAAE,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;YACzG,MAAM,IAAI,GAAgB,SAAS,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;gBACpD,EAAE,EAAE,GAAG,CAAC,EAAE;gBACV,IAAI,EAAE,CAAC,GAAG,CAAC;gBACX,KAAK,EAAE,GAAG,CAAC,KAAK;gBAChB,MAAM,EAAE,OAAO;aACf,CAAC,CAAC,CAAC;YACJ,KAAK,MAAM,GAAG,IAAI,SAAS,EAAE,CAAC;gBAC7B,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,EAAE,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,SAAS,EAAE,GAAG,CAAC,SAAS,EAAE,OAAO,EAAE,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC;YACvF,CAAC;YACD,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACjB,cAAc,CAAC,KAAK,GAAG,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,QAAQ,EAAE,IAAI,CAAC,MAAM,EAAE,CAAC;QACrF,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACZ,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC3D,cAAc,CAAC,KAAK,GAAG,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;QAC3E,CAAC;IAAA,CACD,CAAC;IAEF,MAAM,IAAI,GAAoB,EAAE,CAAC;IACjC,IAAI,IAAI,KAAK,SAAS,IAAI,IAAI,KAAK,QAAQ;QAAE,IAAI,CAAC,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC;IACrE,IAAI,IAAI,KAAK,UAAU,IAAI,IAAI,KAAK,QAAQ;QAAE,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC;IACpE,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACxB,IAAI,MAAM,EAAE,OAAO;QAAE,MAAM,IAAI,KAAK,CAAC,mBAAmB,CAAC,CAAC;IAC1D,wEAAwE;IACxE,iCAAiC;IACjC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,MAAM,MAAM,CAAC,CAAC,CAAC,IAAI,IAAI,KAAK,CAAC,sCAAsC,CAAC,CAAC;IAE7F,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,YAAY,CAAC,CAAC;IAC1D,IAAI,UAAU,GAAqB,EAAE,CAAC;IACtC,KAAK,MAAM,GAAG,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,IAAI,GAAG,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAC/B,IAAI,IAAI;YAAE,UAAU,CAAC,IAAI,CAAC,EAAE,GAAG,GAAG,EAAE,GAAG,IAAI,EAAE,CAAC,CAAC;IAChD,CAAC;IAED,IAAI,UAAiC,CAAC;IACtC,IAAI,OAAO,CAAC,MAAM,KAAK,KAAK,EAAE,CAAC;QAC9B,MAAM,QAAQ,GAAG,gBAAgB,CAAC,KAAK,EAAE,UAAU,EAAE,GAAG,CAAC,CAAC;QAC1D,UAAU,GAAG,EAAE,OAAO,EAAE,IAAI,EAAE,cAAc,EAAE,UAAU,CAAC,MAAM,EAAE,SAAS,EAAE,QAAQ,CAAC,SAAS,EAAE,CAAC;QACjG,UAAU,GAAG,QAAQ,CAAC,UAAU,CAAC;IAClC,CAAC;IACD,UAAU,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;IAExC,OAAO;QACN,UAAU;QACV,YAAY,EAAE,IAAI;QAClB,cAAc,EAAE,UAAU,CAAC,cAAc;QACzC,UAAU,EAAE,KAAK,EAAE,KAAK,KAAK,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,EAAE,KAAK,KAAK,UAAU,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,aAAa;QACzG,UAAU,EAAE,cAAc;QAC1B,QAAQ,EAAE,KAAK,EAAE,KAAK,KAAK,UAAU,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,SAAS;QAC5F,IAAI;QACJ,MAAM,EAAE,UAAU;KAClB,CAAC;AAAA,CACF;AAED,MAAM,CAAC,KAAK,UAAU,SAAS,CAAC,OAAyB,EAA4B;IACpF,MAAM,SAAS,GAAG,MAAM,kBAAkB,CAAC,OAAO,CAAC,CAAC;IAEpD,MAAM,SAAS,GAAG,eAAe,CAAC,SAAS,CAAC,UAAU,EAAE,EAAE,GAAG,EAAE,OAAO,CAAC,GAAG,EAAE,WAAW,EAAE,OAAO,CAAC,WAAW,EAAE,CAAC,CAAC;IAEhH,gBAAgB,CAAC,OAAO,CAAC,GAAG,EAAE;QAC7B,WAAW,EAAE,IAAI,CAAC,GAAG,EAAE;QACvB,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,aAAa,EAAE,OAAO,CAAC,IAAI,IAAI,MAAM;QACrC,YAAY,EAAE,SAAS,CAAC,YAAY;QACpC,cAAc,EAAE,SAAS,CAAC,cAAc;QACxC,UAAU,EAAE,SAAS,CAAC,UAAU;QAChC,IAAI,EAAE,SAAS,CAAC,YAAY,KAAK,QAAQ,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS;QACtE,UAAU,EAAE,SAAS,CAAC,UAAU;QAChC,KAAK,EAAE,SAAS,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC,CAAC,EAAE,EAAE,EAAE,QAAQ,EAAE,KAAK,EAAE,SAAS,EAAE,CAAC,CAAC;QAC7G,MAAM,EAAE,SAAS,CAAC,MAAM;KACxB,CAAC,CAAC;IAEH,OAAO;QACN,IAAI,EAAE,SAAS,CAAC,IAAI;QACpB,YAAY,EAAE,SAAS,CAAC,YAAY;QACpC,cAAc,EAAE,SAAS,CAAC,cAAc;QACxC,WAAW,EAAE,SAAS,CAAC,UAAU,CAAC,MAAM;QACxC,QAAQ,EAAE,SAAS,CAAC,QAAQ;KAC5B,CAAC;AAAA,CACF","sourcesContent":["/**\n * Hybrid search orchestrator: resolve mode, run retrievers in parallel, fuse\n * by rank, expand within budget, trace everything\n * (docs/hybrid-retrieval-design.md).\n *\n * Single-retriever modes flow through the same pipeline — rrfFuse over one\n * list preserves its order — so lexical, semantic, and hybrid all produce the\n * same result shape and the same trace record.\n *\n * `retrieveCandidates` is the candidate-level core (also used by the eval\n * harness, which needs forced modes, a configurable `k`, and no trace\n * pollution); `runSearch` wraps it with span expansion and tracing for the\n * tool.\n */\n\nimport type { EmbsearchService } from \"../embsearch/embsearch-service.js\";\nimport { adaptGrepHits, type ChunkLookup } from \"./adapter.js\";\nimport { assembleContext } from \"./context-assembler.js\";\nimport { runLexicalRetriever } from \"./lexical-retriever.js\";\nimport { resolveSearchMode } from \"./mode.js\";\nimport { rerankCandidates } from \"./rerank.js\";\nimport { DEFAULT_RRF_K, rrfFuse } from \"./rrf.js\";\nimport { writeSearchTrace } from \"./trace.js\";\nimport type { CandidateSpan, FusedCandidate, RankedHit, ResolvedSearchMode, SearchMode, SearchTrace } from \"./types.js\";\n\n/** Raw grep line-hits fetched per query (pre-collapse). */\nconst LEXICAL_MATCH_LIMIT = 200;\n/** Adapted lexical candidates entering fusion. The eval gate showed the\n *  uncapped lexical tail diluting hybrid below plain semantic: lexical\n *  precision is front-loaded by the adapter's term-evidence ranking, while\n *  RRF weighs a rank-30 lexical candidate like a rank-30 embedding hit. */\nconst LEXICAL_FUSION_CAP = 20;\n/** Embedding hits fetched per query — deep enough for fusion to matter. */\nconst EMBED_TOP_K = 50;\n/** Fused candidates kept for reranking / final slicing. */\nconst FUSED_WINDOW = 50;\n\nexport interface RetrieveOptions {\n\tcwd: string;\n\tquery: string;\n\tmode?: SearchMode;\n\t/** Optional glob filter applied to file paths. */\n\tglob?: string;\n\t/** Maximum fused candidates returned. */\n\tlimit?: number;\n\t/** RRF constant override (eval harness sweeps this). Default: {@link DEFAULT_RRF_K}. */\n\trrfK?: number;\n\t/** Rerank the fused top-50 before slicing to `limit`. Default: true. */\n\trerank?: boolean;\n\tservice?: EmbsearchService;\n\tsignal?: AbortSignal;\n}\n\nexport interface RetrieveResult {\n\tcandidates: FusedCandidate[];\n\tresolvedMode: ResolvedSearchMode;\n\tdegradedReason?: string;\n\tindexPhase: SearchTrace[\"indexPhase\"];\n\tretrievers: SearchTrace[\"retrievers\"];\n\t/** Set while the embedding index is still building. */\n\tindexing?: { done: number; total: number };\n\trrfK: number;\n\trerank?: SearchTrace[\"rerank\"];\n}\n\nexport interface RunSearchOptions extends RetrieveOptions {\n\t/** Approximate token budget for the result text. */\n\ttokenBudget?: number;\n}\n\nexport interface RunSearchResult {\n\ttext: string;\n\tresolvedMode: ResolvedSearchMode;\n\tdegradedReason?: string;\n\tresultCount: number;\n\t/** Set while the embedding index is still building. */\n\tindexing?: { done: number; total: number };\n}\n\nfunction normalizeSearchGlob(glob: string | undefined): string | undefined {\n\tif (!glob) return undefined;\n\t// Match fd/rg semantics: a slash-containing glob is anchored anywhere in\n\t// the tree, so prepend \"**/\" unless it already starts with a slash or \"**/\".\n\tif (glob.includes(\"/\") && !glob.startsWith(\"/\") && !glob.startsWith(\"**/\")) {\n\t\treturn `**/${glob}`;\n\t}\n\treturn glob;\n}\n\nexport async function retrieveCandidates(options: RetrieveOptions): Promise<RetrieveResult> {\n\tconst { cwd, query, service, signal } = options;\n\tconst glob = normalizeSearchGlob(options.glob);\n\tconst requestedMode = options.mode ?? \"auto\";\n\tconst limit = Math.max(1, options.limit ?? 10);\n\tconst rrfK = options.rrfK ?? DEFAULT_RRF_K;\n\n\tconst state = service?.getState();\n\tconst embedAvailable = service?.isAvailable() ?? false;\n\tconst embedUnavailableReason =\n\t\tstate === undefined\n\t\t\t? \"semantic index is not enabled\"\n\t\t\t: state.phase === \"unavailable\" || state.phase === \"skipped\"\n\t\t\t\t? state.reason\n\t\t\t\t: state.phase === \"idle\"\n\t\t\t\t\t? \"semantic index has not started\"\n\t\t\t\t\t: undefined;\n\n\tconst resolution = resolveSearchMode(query, requestedMode, embedAvailable, embedUnavailableReason);\n\tconst mode = resolution.mode;\n\n\t// Map lexical hits onto indexed chunk ids whenever the sidecar is usable,\n\t// even in lexical-only mode, so identities line up across modes.\n\tconst lookupChunk: ChunkLookup | undefined = embedAvailable\n\t\t? (rel, line) => service!.findEnclosingChunk(rel, line)\n\t\t: undefined;\n\n\tconst spans = new Map<string, CandidateSpan>();\n\tconst lists: RankedHit[][] = [];\n\tconst retrieverStats: SearchTrace[\"retrievers\"] = {};\n\tconst errors: Error[] = [];\n\n\tconst runLexical = async (): Promise<void> => {\n\t\tconst startedMs = Date.now();\n\t\ttry {\n\t\t\tconst lineHits = await runLexicalRetriever({ cwd, query, limit: LEXICAL_MATCH_LIMIT, glob, signal });\n\t\t\tconst adapted = adaptGrepHits(lineHits, lookupChunk);\n\t\t\t// In single-retriever lexical mode the full list is the result; in\n\t\t\t// hybrid, only the front-loaded head is trustworthy enough to vote.\n\t\t\tconst hits = mode === \"hybrid\" ? adapted.hits.slice(0, LEXICAL_FUSION_CAP) : adapted.hits;\n\t\t\tfor (const [id, span] of adapted.spans) if (!spans.has(id)) spans.set(id, span);\n\t\t\tlists.push(hits);\n\t\t\tretrieverStats.grep = { latencyMs: Date.now() - startedMs, hitCount: hits.length };\n\t\t} catch (e) {\n\t\t\terrors.push(e instanceof Error ? e : new Error(String(e)));\n\t\t\tretrieverStats.grep = { latencyMs: Date.now() - startedMs, hitCount: 0 };\n\t\t}\n\t};\n\n\tconst runEmbed = async (): Promise<void> => {\n\t\tconst startedMs = Date.now();\n\t\ttry {\n\t\t\t// The flat index pads top-k with whatever exists; with the cosine\n\t\t\t// metric the store uses, score <= 0 means \"no relation at all\", so\n\t\t\t// those padding hits would cast RRF votes on pure noise.\n\t\t\tconst chunkHits = (await service!.searchChunks(query, EMBED_TOP_K, glob)).filter((hit) => hit.score > 0);\n\t\t\tconst hits: RankedHit[] = chunkHits.map((hit, i) => ({\n\t\t\t\tid: hit.id,\n\t\t\t\trank: i + 1,\n\t\t\t\tscore: hit.score,\n\t\t\t\tsource: \"embed\",\n\t\t\t}));\n\t\t\tfor (const hit of chunkHits) {\n\t\t\t\tspans.set(hit.id, { path: hit.path, startLine: hit.startLine, endLine: hit.endLine });\n\t\t\t}\n\t\t\tlists.push(hits);\n\t\t\tretrieverStats.embed = { latencyMs: Date.now() - startedMs, hitCount: hits.length };\n\t\t} catch (e) {\n\t\t\terrors.push(e instanceof Error ? e : new Error(String(e)));\n\t\t\tretrieverStats.embed = { latencyMs: Date.now() - startedMs, hitCount: 0 };\n\t\t}\n\t};\n\n\tconst runs: Promise<void>[] = [];\n\tif (mode === \"lexical\" || mode === \"hybrid\") runs.push(runLexical());\n\tif (mode === \"semantic\" || mode === \"hybrid\") runs.push(runEmbed());\n\tawait Promise.all(runs);\n\tif (signal?.aborted) throw new Error(\"Operation aborted\");\n\t// A partial failure in hybrid degrades to whichever retriever survived;\n\t// only a total loss is an error.\n\tif (lists.length === 0) throw errors[0] ?? new Error(\"search produced no retriever results\");\n\n\tconst fused = rrfFuse(lists, rrfK).slice(0, FUSED_WINDOW);\n\tlet candidates: FusedCandidate[] = [];\n\tfor (const hit of fused) {\n\t\tconst span = spans.get(hit.id);\n\t\tif (span) candidates.push({ ...hit, ...span });\n\t}\n\n\tlet rerankInfo: SearchTrace[\"rerank\"];\n\tif (options.rerank !== false) {\n\t\tconst reranked = rerankCandidates(query, candidates, cwd);\n\t\trerankInfo = { applied: true, candidateCount: candidates.length, latencyMs: reranked.latencyMs };\n\t\tcandidates = reranked.candidates;\n\t}\n\tcandidates = candidates.slice(0, limit);\n\n\treturn {\n\t\tcandidates,\n\t\tresolvedMode: mode,\n\t\tdegradedReason: resolution.degradedReason,\n\t\tindexPhase: state?.phase === \"ready\" ? \"ready\" : state?.phase === \"indexing\" ? \"indexing\" : \"unavailable\",\n\t\tretrievers: retrieverStats,\n\t\tindexing: state?.phase === \"indexing\" ? { done: state.done, total: state.total } : undefined,\n\t\trrfK,\n\t\trerank: rerankInfo,\n\t};\n}\n\nexport async function runSearch(options: RunSearchOptions): Promise<RunSearchResult> {\n\tconst retrieved = await retrieveCandidates(options);\n\n\tconst assembled = assembleContext(retrieved.candidates, { cwd: options.cwd, tokenBudget: options.tokenBudget });\n\n\twriteSearchTrace(options.cwd, {\n\t\ttimestampMs: Date.now(),\n\t\tquery: options.query,\n\t\trequestedMode: options.mode ?? \"auto\",\n\t\tresolvedMode: retrieved.resolvedMode,\n\t\tdegradedReason: retrieved.degradedReason,\n\t\tindexPhase: retrieved.indexPhase,\n\t\trrfK: retrieved.resolvedMode === \"hybrid\" ? retrieved.rrfK : undefined,\n\t\tretrievers: retrieved.retrievers,\n\t\tfused: retrieved.candidates.map(({ id, rrfScore, ranks, rawScores }) => ({ id, rrfScore, ranks, rawScores })),\n\t\trerank: retrieved.rerank,\n\t});\n\n\treturn {\n\t\ttext: assembled.text,\n\t\tresolvedMode: retrieved.resolvedMode,\n\t\tdegradedReason: retrieved.degradedReason,\n\t\tresultCount: retrieved.candidates.length,\n\t\tindexing: retrieved.indexing,\n\t};\n}\n"]}

package/dist/core/search/lexical-retriever.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"lexical-retriever.d.ts","sourceRoot":"","sources":["../../../src/core/search/lexical-retriever.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAQH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAWhD,MAAM,WAAW,gBAAgB;IAChC,8BAA8B;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,wEAAwE;IACxE,KAAK,EAAE,MAAM,EAAE,CAAC;CAChB;AAED;;;;GAIG;AACH,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,MAAM,GAAG,gBAAgB,GAAG,SAAS,CAgBjF;AAED,0DAA0D;AAC1D,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAErE;AAUD;;;GAGG;AACH,MAAM,WAAW,iBAAiB;IACjC,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,WAAW,CAAC;CACrB;AAiBD,wBAAsB,mBAAmB,CAAC,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC,CA4F5F","sourcesContent":["/**\n * Internal lexical retriever for hybrid search.\n *\n * This is not the grep *tool* — it is the lexical recall backend: it turns a\n * natural query into a ripgrep pattern, streams matches, and returns bare\n * `rel:line` hits for the grep→chunk adapter. rg drives the fast path; the\n * pure-JS nativeGrep fallback keeps restricted environments working, same as\n * the grep tool.\n */\n\nimport { createInterface } from \"node:readline\";\nimport { spawn } from \"child_process\";\nimport { readFileSync } from \"fs\";\nimport path from \"path\";\nimport { ensureTool } from \"../../utils/tools-manager.js\";\nimport { isNativeSearchForced, nativeGrep } from \"../tools/native-search.js\";\nimport type { GrepLineHit } from \"./adapter.js\";\n\n/** Terms considered per query (longest first) when building the pattern. */\nconst MAX_TERMS = 4;\n/** Minimum token length worth matching on. */\nconst MIN_TERM_LENGTH = 3;\n\nfunction escapeRegExp(value: string): string {\n\treturn value.replace(/[.*+?^${}()|[\\]\\\\]/g, \"\\\\$&\");\n}\n\nexport interface LexicalQueryPlan {\n\t/** rg-ready regex pattern. */\n\tpattern: string;\n\t/** Raw (unescaped, lowercased) terms, for per-line term attribution. */\n\tterms: string[];\n}\n\n/**\n * Build the retrieval plan for a query: a quoted segment is searched\n * verbatim; otherwise the longest few identifier-ish tokens are OR-ed\n * together. Returns undefined when the query yields nothing searchable.\n */\nexport function buildLexicalQueryPlan(query: string): LexicalQueryPlan | undefined {\n\tconst quoted = [...query.matchAll(/[\"'`]([^\"'`]+)[\"'`]/g)]\n\t\t.map((m) => m[1].trim())\n\t\t.filter((s) => s.length > 0)\n\t\t.sort((a, b) => b.length - a.length)[0];\n\tif (quoted) return { pattern: escapeRegExp(quoted), terms: [quoted.toLowerCase()] };\n\n\tconst tokens = [...new Set(query.match(/[A-Za-z0-9_$][\\w$.-]*/g) ?? [])]\n\t\t.filter((t) => t.length >= MIN_TERM_LENGTH)\n\t\t.sort((a, b) => b.length - a.length || a.localeCompare(b))\n\t\t.slice(0, MAX_TERMS);\n\tif (tokens.length === 0) {\n\t\tconst trimmed = query.trim();\n\t\treturn trimmed ? { pattern: escapeRegExp(trimmed), terms: [trimmed.toLowerCase()] } : undefined;\n\t}\n\treturn { pattern: tokens.map(escapeRegExp).join(\"|\"), terms: tokens.map((t) => t.toLowerCase()) };\n}\n\n/** Pattern-only view of {@link buildLexicalQueryPlan}. */\nexport function buildLexicalPattern(query: string): string | undefined {\n\treturn buildLexicalQueryPlan(query)?.pattern;\n}\n\n/** Which plan terms appear on a matched line (retrieval is case-insensitive,\n *  so attribution is too). */\nfunction termsOnLine(plan: LexicalQueryPlan, lineText: string | undefined): string[] {\n\tif (!lineText) return [];\n\tconst lower = lineText.toLowerCase();\n\treturn plan.terms.filter((t) => lower.includes(t));\n}\n\n/**\n * Run lexical retrieval over `cwd`, returning up to `limit` line-hits in\n * output order with POSIX repo-relative paths.\n */\nexport interface RunLexicalOptions {\n\tcwd: string;\n\tquery: string;\n\tlimit: number;\n\tglob?: string;\n\tsignal?: AbortSignal;\n}\n\n/**\n * Run the lexical retriever for the search tool. Optional glob filter scopes\n * file paths (slashless matches basename anywhere, slash patterns match the\n * full repo-relative path).\n */\nfunction normalizeSearchGlob(glob: string | undefined): string | undefined {\n\tif (!glob) return undefined;\n\t// Match fd/rg semantics: a slash-containing glob is anchored anywhere in\n\t// the tree, so prepend \"**/\" unless it already starts with a slash or \"**/\".\n\tif (glob.includes(\"/\") && !glob.startsWith(\"/\") && !glob.startsWith(\"**/\")) {\n\t\treturn `**/${glob}`;\n\t}\n\treturn glob;\n}\n\nexport async function runLexicalRetriever(options: RunLexicalOptions): Promise<GrepLineHit[]> {\n\tconst { cwd, query, limit, glob: rawGlob, signal } = options;\n\tconst glob = normalizeSearchGlob(rawGlob);\n\tconst plan = buildLexicalQueryPlan(query);\n\tif (!plan) return [];\n\tconst { pattern } = plan;\n\n\tconst toRel = (filePath: string): string => {\n\t\tconst rel = path.relative(cwd, filePath);\n\t\treturn (rel && !rel.startsWith(\"..\") ? rel : filePath).replace(/\\\\/g, \"/\");\n\t};\n\n\tconst rgPath = isNativeSearchForced() ? undefined : await ensureTool(\"rg\", true);\n\tif (!rgPath) {\n\t\tconst result = await nativeGrep(cwd, {\n\t\t\tpattern,\n\t\t\tisDirectory: true,\n\t\t\tignoreCase: true,\n\t\t\tlimit,\n\t\t\tglob,\n\t\t\tsignal,\n\t\t\treadFile: (p) => readFileSync(p, \"utf-8\"),\n\t\t});\n\t\treturn result.matches.map((m) => ({\n\t\t\trel: toRel(m.filePath),\n\t\t\tline: m.lineNumber,\n\t\t\tterms: termsOnLine(plan, m.lineText),\n\t\t}));\n\t}\n\n\treturn new Promise<GrepLineHit[]>((resolve, reject) => {\n\t\t// --sort path forces a deterministic (single-threaded) walk: with the\n\t\t// match cap truncating the stream, a parallel walk would return a\n\t\t// different hit subset per run — \"same query, different context\".\n\t\tconst args = [\"--json\", \"--line-number\", \"--color=never\", \"--hidden\", \"--ignore-case\", \"--sort\", \"path\"];\n\t\tif (glob) args.push(\"--glob\", glob);\n\t\targs.push(\"--\", pattern, cwd);\n\t\tconst child = spawn(rgPath, args, { stdio: [\"ignore\", \"pipe\", \"pipe\"] });\n\t\tconst rl = createInterface({ input: child.stdout });\n\t\tconst hits: GrepLineHit[] = [];\n\t\tlet stderr = \"\";\n\t\tlet killedDueToLimit = false;\n\t\tlet aborted = false;\n\n\t\tconst onAbort = () => {\n\t\t\taborted = true;\n\t\t\tif (!child.killed) child.kill();\n\t\t};\n\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\t\tchild.stderr?.on(\"data\", (chunk) => {\n\t\t\tstderr += chunk.toString();\n\t\t});\n\n\t\trl.on(\"line\", (line) => {\n\t\t\tif (!line.trim() || hits.length >= limit) return;\n\t\t\tlet event: any;\n\t\t\ttry {\n\t\t\t\tevent = JSON.parse(line);\n\t\t\t} catch {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (event.type !== \"match\") return;\n\t\t\tconst filePath = event.data?.path?.text;\n\t\t\tconst lineNumber = event.data?.line_number;\n\t\t\tif (filePath && typeof lineNumber === \"number\") {\n\t\t\t\thits.push({ rel: toRel(filePath), line: lineNumber, terms: termsOnLine(plan, event.data?.lines?.text) });\n\t\t\t}\n\t\t\tif (hits.length >= limit && !child.killed) {\n\t\t\t\tkilledDueToLimit = true;\n\t\t\t\tchild.kill();\n\t\t\t}\n\t\t});\n\n\t\tchild.on(\"error\", (error) => {\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\treject(new Error(`Failed to run ripgrep: ${error.message}`));\n\t\t});\n\t\tchild.on(\"close\", (code) => {\n\t\t\trl.close();\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\tif (aborted) {\n\t\t\t\treject(new Error(\"Operation aborted\"));\n\t\t\t\treturn;\n\t\t\t}\n\t\t\t// rg exits 1 on \"no matches\" — that is a valid empty result.\n\t\t\tif (!killedDueToLimit && code !== 0 && code !== 1) {\n\t\t\t\treject(new Error(stderr.trim() || `ripgrep exited with code ${code}`));\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tresolve(hits);\n\t\t});\n\t});\n}\n"]}
+{"version":3,"file":"lexical-retriever.d.ts","sourceRoot":"","sources":["../../../src/core/search/lexical-retriever.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAQH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAC;AAWhD,MAAM,WAAW,gBAAgB;IAChC,8BAA8B;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,wEAAwE;IACxE,KAAK,EAAE,MAAM,EAAE,CAAC;CAChB;AAED;;;;GAIG;AACH,wBAAgB,qBAAqB,CAAC,KAAK,EAAE,MAAM,GAAG,gBAAgB,GAAG,SAAS,CAgBjF;AAED,0DAA0D;AAC1D,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAErE;AAUD;;;GAGG;AACH,MAAM,WAAW,iBAAiB;IACjC,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,MAAM,CAAC,EAAE,WAAW,CAAC;CACrB;AAiBD,wBAAsB,mBAAmB,CAAC,OAAO,EAAE,iBAAiB,GAAG,OAAO,CAAC,WAAW,EAAE,CAAC,CAyG5F","sourcesContent":["/**\n * Internal lexical retriever for hybrid search.\n *\n * This is not the grep *tool* — it is the lexical recall backend: it turns a\n * natural query into a ripgrep pattern, streams matches, and returns bare\n * `rel:line` hits for the grep→chunk adapter. rg drives the fast path; the\n * pure-JS nativeGrep fallback keeps restricted environments working, same as\n * the grep tool.\n */\n\nimport { createInterface } from \"node:readline\";\nimport { spawn } from \"child_process\";\nimport { readFileSync } from \"fs\";\nimport path from \"path\";\nimport { ensureTool } from \"../../utils/tools-manager.js\";\nimport { isNativeSearchForced, nativeGrep } from \"../tools/native-search.js\";\nimport type { GrepLineHit } from \"./adapter.js\";\n\n/** Terms considered per query (longest first) when building the pattern. */\nconst MAX_TERMS = 4;\n/** Minimum token length worth matching on. */\nconst MIN_TERM_LENGTH = 3;\n\nfunction escapeRegExp(value: string): string {\n\treturn value.replace(/[.*+?^${}()|[\\]\\\\]/g, \"\\\\$&\");\n}\n\nexport interface LexicalQueryPlan {\n\t/** rg-ready regex pattern. */\n\tpattern: string;\n\t/** Raw (unescaped, lowercased) terms, for per-line term attribution. */\n\tterms: string[];\n}\n\n/**\n * Build the retrieval plan for a query: a quoted segment is searched\n * verbatim; otherwise the longest few identifier-ish tokens are OR-ed\n * together. Returns undefined when the query yields nothing searchable.\n */\nexport function buildLexicalQueryPlan(query: string): LexicalQueryPlan | undefined {\n\tconst quoted = [...query.matchAll(/[\"'`]([^\"'`]+)[\"'`]/g)]\n\t\t.map((m) => m[1].trim())\n\t\t.filter((s) => s.length > 0)\n\t\t.sort((a, b) => b.length - a.length)[0];\n\tif (quoted) return { pattern: escapeRegExp(quoted), terms: [quoted.toLowerCase()] };\n\n\tconst tokens = [...new Set(query.match(/[A-Za-z0-9_$][\\w$.-]*/g) ?? [])]\n\t\t.filter((t) => t.length >= MIN_TERM_LENGTH)\n\t\t.sort((a, b) => b.length - a.length || a.localeCompare(b))\n\t\t.slice(0, MAX_TERMS);\n\tif (tokens.length === 0) {\n\t\tconst trimmed = query.trim();\n\t\treturn trimmed ? { pattern: escapeRegExp(trimmed), terms: [trimmed.toLowerCase()] } : undefined;\n\t}\n\treturn { pattern: tokens.map(escapeRegExp).join(\"|\"), terms: tokens.map((t) => t.toLowerCase()) };\n}\n\n/** Pattern-only view of {@link buildLexicalQueryPlan}. */\nexport function buildLexicalPattern(query: string): string | undefined {\n\treturn buildLexicalQueryPlan(query)?.pattern;\n}\n\n/** Which plan terms appear on a matched line (retrieval is case-insensitive,\n *  so attribution is too). */\nfunction termsOnLine(plan: LexicalQueryPlan, lineText: string | undefined): string[] {\n\tif (!lineText) return [];\n\tconst lower = lineText.toLowerCase();\n\treturn plan.terms.filter((t) => lower.includes(t));\n}\n\n/**\n * Run lexical retrieval over `cwd`, returning up to `limit` line-hits in\n * output order with POSIX repo-relative paths.\n */\nexport interface RunLexicalOptions {\n\tcwd: string;\n\tquery: string;\n\tlimit: number;\n\tglob?: string;\n\tsignal?: AbortSignal;\n}\n\n/**\n * Run the lexical retriever for the search tool. Optional glob filter scopes\n * file paths (slashless matches basename anywhere, slash patterns match the\n * full repo-relative path).\n */\nfunction normalizeSearchGlob(glob: string | undefined): string | undefined {\n\tif (!glob) return undefined;\n\t// Match fd/rg semantics: a slash-containing glob is anchored anywhere in\n\t// the tree, so prepend \"**/\" unless it already starts with a slash or \"**/\".\n\tif (glob.includes(\"/\") && !glob.startsWith(\"/\") && !glob.startsWith(\"**/\")) {\n\t\treturn `**/${glob}`;\n\t}\n\treturn glob;\n}\n\nexport async function runLexicalRetriever(options: RunLexicalOptions): Promise<GrepLineHit[]> {\n\tconst { cwd, query, limit, glob: rawGlob, signal } = options;\n\tconst glob = normalizeSearchGlob(rawGlob);\n\tconst plan = buildLexicalQueryPlan(query);\n\tif (!plan) return [];\n\tconst { pattern } = plan;\n\n\tconst toRel = (filePath: string): string => {\n\t\tconst rel = path.relative(cwd, filePath);\n\t\treturn (rel && !rel.startsWith(\"..\") ? rel : filePath).replace(/\\\\/g, \"/\");\n\t};\n\n\tconst rgPath = isNativeSearchForced() ? undefined : await ensureTool(\"rg\", true);\n\tif (!rgPath) {\n\t\tconst result = await nativeGrep(cwd, {\n\t\t\tpattern,\n\t\t\tisDirectory: true,\n\t\t\tignoreCase: true,\n\t\t\tlimit,\n\t\t\tglob,\n\t\t\tsignal,\n\t\t\treadFile: (p) => readFileSync(p, \"utf-8\"),\n\t\t});\n\t\treturn result.matches.map((m) => ({\n\t\t\trel: toRel(m.filePath),\n\t\t\tline: m.lineNumber,\n\t\t\tterms: termsOnLine(plan, m.lineText),\n\t\t}));\n\t}\n\n\treturn new Promise<GrepLineHit[]>((resolve, reject) => {\n\t\t// --sort path forces a deterministic (single-threaded) walk: with the\n\t\t// match cap truncating the stream, a parallel walk would return a\n\t\t// different hit subset per run — \"same query, different context\".\n\t\t// `!.git` because --hidden would otherwise search .git contents;\n\t\t// --no-require-git so .gitignore is honored outside git repos too.\n\t\tconst args = [\n\t\t\t\"--json\",\n\t\t\t\"--line-number\",\n\t\t\t\"--color=never\",\n\t\t\t\"--hidden\",\n\t\t\t\"--no-require-git\",\n\t\t\t\"--ignore-case\",\n\t\t\t\"--sort\",\n\t\t\t\"path\",\n\t\t\t\"--glob\",\n\t\t\t\"!**/.git/**\",\n\t\t];\n\t\tif (glob) args.push(\"--glob\", glob);\n\t\targs.push(\"--\", pattern, cwd);\n\t\tconst child = spawn(rgPath, args, { stdio: [\"ignore\", \"pipe\", \"pipe\"] });\n\t\tconst rl = createInterface({ input: child.stdout });\n\t\tconst hits: GrepLineHit[] = [];\n\t\tlet stderr = \"\";\n\t\tlet killedDueToLimit = false;\n\t\tlet aborted = false;\n\n\t\tconst onAbort = () => {\n\t\t\taborted = true;\n\t\t\tif (!child.killed) child.kill();\n\t\t};\n\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\t\tchild.stderr?.on(\"data\", (chunk) => {\n\t\t\tstderr += chunk.toString();\n\t\t});\n\n\t\trl.on(\"line\", (line) => {\n\t\t\tif (!line.trim() || hits.length >= limit) return;\n\t\t\tlet event: any;\n\t\t\ttry {\n\t\t\t\tevent = JSON.parse(line);\n\t\t\t} catch {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (event.type !== \"match\") return;\n\t\t\tconst filePath = event.data?.path?.text;\n\t\t\tconst lineNumber = event.data?.line_number;\n\t\t\tif (filePath && typeof lineNumber === \"number\") {\n\t\t\t\thits.push({ rel: toRel(filePath), line: lineNumber, terms: termsOnLine(plan, event.data?.lines?.text) });\n\t\t\t}\n\t\t\tif (hits.length >= limit && !child.killed) {\n\t\t\t\tkilledDueToLimit = true;\n\t\t\t\tchild.kill();\n\t\t\t}\n\t\t});\n\n\t\tchild.on(\"error\", (error) => {\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\treject(new Error(`Failed to run ripgrep: ${error.message}`));\n\t\t});\n\t\tchild.on(\"close\", (code) => {\n\t\t\trl.close();\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\tif (aborted) {\n\t\t\t\treject(new Error(\"Operation aborted\"));\n\t\t\t\treturn;\n\t\t\t}\n\t\t\t// rg exits 1 on \"no matches\" — that is a valid empty result.\n\t\t\tif (!killedDueToLimit && code !== 0 && code !== 1) {\n\t\t\t\treject(new Error(stderr.trim() || `ripgrep exited with code ${code}`));\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tresolve(hits);\n\t\t});\n\t});\n}\n"]}

package/dist/core/search/lexical-retriever.js

@@ -101,7 +101,20 @@ export async function runLexicalRetriever(options) {
         // --sort path forces a deterministic (single-threaded) walk: with the
         // match cap truncating the stream, a parallel walk would return a
         // different hit subset per run — "same query, different context".
-        const args = ["--json", "--line-number", "--color=never", "--hidden", "--ignore-case", "--sort", "path"];
+        // `!.git` because --hidden would otherwise search .git contents;
+        // --no-require-git so .gitignore is honored outside git repos too.
+        const args = [
+            "--json",
+            "--line-number",
+            "--color=never",
+            "--hidden",
+            "--no-require-git",
+            "--ignore-case",
+            "--sort",
+            "path",
+            "--glob",
+            "!**/.git/**",
+        ];
         if (glob)
             args.push("--glob", glob);
         args.push("--", pattern, cwd);

package/dist/core/search/lexical-retriever.js.map

@@ -1 +1 @@
-{"version":3,"file":"lexical-retriever.js","sourceRoot":"","sources":["../../../src/core/search/lexical-retriever.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAChD,OAAO,EAAE,KAAK,EAAE,MAAM,eAAe,CAAC;AACtC,OAAO,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;AAClC,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAE,UAAU,EAAE,MAAM,8BAA8B,CAAC;AAC1D,OAAO,EAAE,oBAAoB,EAAE,UAAU,EAAE,MAAM,2BAA2B,CAAC;AAG7E,4EAA4E;AAC5E,MAAM,SAAS,GAAG,CAAC,CAAC;AACpB,8CAA8C;AAC9C,MAAM,eAAe,GAAG,CAAC,CAAC;AAE1B,SAAS,YAAY,CAAC,KAAa,EAAU;IAC5C,OAAO,KAAK,CAAC,OAAO,CAAC,qBAAqB,EAAE,MAAM,CAAC,CAAC;AAAA,CACpD;AASD;;;;GAIG;AACH,MAAM,UAAU,qBAAqB,CAAC,KAAa,EAAgC;IAClF,MAAM,MAAM,GAAG,CAAC,GAAG,KAAK,CAAC,QAAQ,CAAC,sBAAsB,CAAC,CAAC;SACxD,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;SACvB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC;SAC3B,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;IACzC,IAAI,MAAM;QAAE,OAAO,EAAE,OAAO,EAAE,YAAY,CAAC,MAAM,CAAC,EAAE,KAAK,EAAE,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC;IAEpF,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,wBAAwB,CAAC,IAAI,EAAE,CAAC,CAAC;SACtE,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,IAAI,eAAe,CAAC;SAC1C,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC;SACzD,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;IACtB,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACzB,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;QAC7B,OAAO,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,YAAY,CAAC,OAAO,CAAC,EAAE,KAAK,EAAE,CAAC,OAAO,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;IACjG,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC;AAAA,CAClG;AAED,0DAA0D;AAC1D,MAAM,UAAU,mBAAmB,CAAC,KAAa,EAAsB;IACtE,OAAO,qBAAqB,CAAC,KAAK,CAAC,EAAE,OAAO,CAAC;AAAA,CAC7C;AAED;8BAC8B;AAC9B,SAAS,WAAW,CAAC,IAAsB,EAAE,QAA4B,EAAY;IACpF,IAAI,CAAC,QAAQ;QAAE,OAAO,EAAE,CAAC;IACzB,MAAM,KAAK,GAAG,QAAQ,CAAC,WAAW,EAAE,CAAC;IACrC,OAAO,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;AAAA,CACnD;AAcD;;;;GAIG;AACH,SAAS,mBAAmB,CAAC,IAAwB,EAAsB;IAC1E,IAAI,CAAC,IAAI;QAAE,OAAO,SAAS,CAAC;IAC5B,yEAAyE;IACzE,6EAA6E;IAC7E,IAAI,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;QAC5E,OAAO,MAAM,IAAI,EAAE,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,MAAM,CAAC,KAAK,UAAU,mBAAmB,CAAC,OAA0B,EAA0B;IAC7F,MAAM,EAAE,GAAG,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC;IAC7D,MAAM,IAAI,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;IAC1C,MAAM,IAAI,GAAG,qBAAqB,CAAC,KAAK,CAAC,CAAC;IAC1C,IAAI,CAAC,IAAI;QAAE,OAAO,EAAE,CAAC;IACrB,MAAM,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC;IAEzB,MAAM,KAAK,GAAG,CAAC,QAAgB,EAAU,EAAE,CAAC;QAC3C,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAC;QACzC,OAAO,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAAA,CAC3E,CAAC;IAEF,MAAM,MAAM,GAAG,oBAAoB,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,UAAU,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;IACjF,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,GAAG,EAAE;YACpC,OAAO;YACP,WAAW,EAAE,IAAI;YACjB,UAAU,EAAE,IAAI;YAChB,KAAK;YACL,IAAI;YACJ,MAAM;YACN,QAAQ,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,CAAC,EAAE,OAAO,CAAC;SACzC,CAAC,CAAC;QACH,OAAO,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACjC,GAAG,EAAE,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;YACtB,IAAI,EAAE,CAAC,CAAC,UAAU;YAClB,KAAK,EAAE,WAAW,CAAC,IAAI,EAAE,CAAC,CAAC,QAAQ,CAAC;SACpC,CAAC,CAAC,CAAC;IACL,CAAC;IAED,OAAO,IAAI,OAAO,CAAgB,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,CAAC;QACtD,sEAAsE;QACtE,kEAAkE;QAClE,oEAAkE;QAClE,MAAM,IAAI,GAAG,CAAC,QAAQ,EAAE,eAAe,EAAE,eAAe,EAAE,UAAU,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC;QACzG,IAAI,IAAI;YAAE,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;QACpC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;QAC9B,MAAM,KAAK,GAAG,KAAK,CAAC,MAAM,EAAE,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC;QACzE,MAAM,EAAE,GAAG,eAAe,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC;QACpD,MAAM,IAAI,GAAkB,EAAE,CAAC;QAC/B,IAAI,MAAM,GAAG,EAAE,CAAC;QAChB,IAAI,gBAAgB,GAAG,KAAK,CAAC;QAC7B,IAAI,OAAO,GAAG,KAAK,CAAC;QAEpB,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;YACrB,OAAO,GAAG,IAAI,CAAC;YACf,IAAI,CAAC,KAAK,CAAC,MAAM;gBAAE,KAAK,CAAC,IAAI,EAAE,CAAC;QAAA,CAChC,CAAC;QACF,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;QAC3D,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC;YACnC,MAAM,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;QAAA,CAC3B,CAAC,CAAC;QAEH,EAAE,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC;YACvB,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,IAAI,CAAC,MAAM,IAAI,KAAK;gBAAE,OAAO;YACjD,IAAI,KAAU,CAAC;YACf,IAAI,CAAC;gBACJ,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;YAC1B,CAAC;YAAC,MAAM,CAAC;gBACR,OAAO;YACR,CAAC;YACD,IAAI,KAAK,CAAC,IAAI,KAAK,OAAO;gBAAE,OAAO;YACnC,MAAM,QAAQ,GAAG,KAAK,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC;YACxC,MAAM,UAAU,GAAG,KAAK,CAAC,IAAI,EAAE,WAAW,CAAC;YAC3C,IAAI,QAAQ,IAAI,OAAO,UAAU,KAAK,QAAQ,EAAE,CAAC;gBAChD,IAAI,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,KAAK,CAAC,QAAQ,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,WAAW,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;YAC1G,CAAC;YACD,IAAI,IAAI,CAAC,MAAM,IAAI,KAAK,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC;gBAC3C,gBAAgB,GAAG,IAAI,CAAC;gBACxB,KAAK,CAAC,IAAI,EAAE,CAAC;YACd,CAAC;QAAA,CACD,CAAC,CAAC;QAEH,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC;YAC5B,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9C,MAAM,CAAC,IAAI,KAAK,CAAC,0BAA0B,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAAA,CAC7D,CAAC,CAAC;QACH,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC;YAC3B,EAAE,CAAC,KAAK,EAAE,CAAC;YACX,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9C,IAAI,OAAO,EAAE,CAAC;gBACb,MAAM,CAAC,IAAI,KAAK,CAAC,mBAAmB,CAAC,CAAC,CAAC;gBACvC,OAAO;YACR,CAAC;YACD,+DAA6D;YAC7D,IAAI,CAAC,gBAAgB,IAAI,IAAI,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,EAAE,CAAC;gBACnD,MAAM,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,IAAI,EAAE,IAAI,4BAA4B,IAAI,EAAE,CAAC,CAAC,CAAC;gBACvE,OAAO;YACR,CAAC;YACD,OAAO,CAAC,IAAI,CAAC,CAAC;QAAA,CACd,CAAC,CAAC;IAAA,CACH,CAAC,CAAC;AAAA,CACH","sourcesContent":["/**\n * Internal lexical retriever for hybrid search.\n *\n * This is not the grep *tool* — it is the lexical recall backend: it turns a\n * natural query into a ripgrep pattern, streams matches, and returns bare\n * `rel:line` hits for the grep→chunk adapter. rg drives the fast path; the\n * pure-JS nativeGrep fallback keeps restricted environments working, same as\n * the grep tool.\n */\n\nimport { createInterface } from \"node:readline\";\nimport { spawn } from \"child_process\";\nimport { readFileSync } from \"fs\";\nimport path from \"path\";\nimport { ensureTool } from \"../../utils/tools-manager.js\";\nimport { isNativeSearchForced, nativeGrep } from \"../tools/native-search.js\";\nimport type { GrepLineHit } from \"./adapter.js\";\n\n/** Terms considered per query (longest first) when building the pattern. */\nconst MAX_TERMS = 4;\n/** Minimum token length worth matching on. */\nconst MIN_TERM_LENGTH = 3;\n\nfunction escapeRegExp(value: string): string {\n\treturn value.replace(/[.*+?^${}()|[\\]\\\\]/g, \"\\\\$&\");\n}\n\nexport interface LexicalQueryPlan {\n\t/** rg-ready regex pattern. */\n\tpattern: string;\n\t/** Raw (unescaped, lowercased) terms, for per-line term attribution. */\n\tterms: string[];\n}\n\n/**\n * Build the retrieval plan for a query: a quoted segment is searched\n * verbatim; otherwise the longest few identifier-ish tokens are OR-ed\n * together. Returns undefined when the query yields nothing searchable.\n */\nexport function buildLexicalQueryPlan(query: string): LexicalQueryPlan | undefined {\n\tconst quoted = [...query.matchAll(/[\"'`]([^\"'`]+)[\"'`]/g)]\n\t\t.map((m) => m[1].trim())\n\t\t.filter((s) => s.length > 0)\n\t\t.sort((a, b) => b.length - a.length)[0];\n\tif (quoted) return { pattern: escapeRegExp(quoted), terms: [quoted.toLowerCase()] };\n\n\tconst tokens = [...new Set(query.match(/[A-Za-z0-9_$][\\w$.-]*/g) ?? [])]\n\t\t.filter((t) => t.length >= MIN_TERM_LENGTH)\n\t\t.sort((a, b) => b.length - a.length || a.localeCompare(b))\n\t\t.slice(0, MAX_TERMS);\n\tif (tokens.length === 0) {\n\t\tconst trimmed = query.trim();\n\t\treturn trimmed ? { pattern: escapeRegExp(trimmed), terms: [trimmed.toLowerCase()] } : undefined;\n\t}\n\treturn { pattern: tokens.map(escapeRegExp).join(\"|\"), terms: tokens.map((t) => t.toLowerCase()) };\n}\n\n/** Pattern-only view of {@link buildLexicalQueryPlan}. */\nexport function buildLexicalPattern(query: string): string | undefined {\n\treturn buildLexicalQueryPlan(query)?.pattern;\n}\n\n/** Which plan terms appear on a matched line (retrieval is case-insensitive,\n *  so attribution is too). */\nfunction termsOnLine(plan: LexicalQueryPlan, lineText: string | undefined): string[] {\n\tif (!lineText) return [];\n\tconst lower = lineText.toLowerCase();\n\treturn plan.terms.filter((t) => lower.includes(t));\n}\n\n/**\n * Run lexical retrieval over `cwd`, returning up to `limit` line-hits in\n * output order with POSIX repo-relative paths.\n */\nexport interface RunLexicalOptions {\n\tcwd: string;\n\tquery: string;\n\tlimit: number;\n\tglob?: string;\n\tsignal?: AbortSignal;\n}\n\n/**\n * Run the lexical retriever for the search tool. Optional glob filter scopes\n * file paths (slashless matches basename anywhere, slash patterns match the\n * full repo-relative path).\n */\nfunction normalizeSearchGlob(glob: string | undefined): string | undefined {\n\tif (!glob) return undefined;\n\t// Match fd/rg semantics: a slash-containing glob is anchored anywhere in\n\t// the tree, so prepend \"**/\" unless it already starts with a slash or \"**/\".\n\tif (glob.includes(\"/\") && !glob.startsWith(\"/\") && !glob.startsWith(\"**/\")) {\n\t\treturn `**/${glob}`;\n\t}\n\treturn glob;\n}\n\nexport async function runLexicalRetriever(options: RunLexicalOptions): Promise<GrepLineHit[]> {\n\tconst { cwd, query, limit, glob: rawGlob, signal } = options;\n\tconst glob = normalizeSearchGlob(rawGlob);\n\tconst plan = buildLexicalQueryPlan(query);\n\tif (!plan) return [];\n\tconst { pattern } = plan;\n\n\tconst toRel = (filePath: string): string => {\n\t\tconst rel = path.relative(cwd, filePath);\n\t\treturn (rel && !rel.startsWith(\"..\") ? rel : filePath).replace(/\\\\/g, \"/\");\n\t};\n\n\tconst rgPath = isNativeSearchForced() ? undefined : await ensureTool(\"rg\", true);\n\tif (!rgPath) {\n\t\tconst result = await nativeGrep(cwd, {\n\t\t\tpattern,\n\t\t\tisDirectory: true,\n\t\t\tignoreCase: true,\n\t\t\tlimit,\n\t\t\tglob,\n\t\t\tsignal,\n\t\t\treadFile: (p) => readFileSync(p, \"utf-8\"),\n\t\t});\n\t\treturn result.matches.map((m) => ({\n\t\t\trel: toRel(m.filePath),\n\t\t\tline: m.lineNumber,\n\t\t\tterms: termsOnLine(plan, m.lineText),\n\t\t}));\n\t}\n\n\treturn new Promise<GrepLineHit[]>((resolve, reject) => {\n\t\t// --sort path forces a deterministic (single-threaded) walk: with the\n\t\t// match cap truncating the stream, a parallel walk would return a\n\t\t// different hit subset per run — \"same query, different context\".\n\t\tconst args = [\"--json\", \"--line-number\", \"--color=never\", \"--hidden\", \"--ignore-case\", \"--sort\", \"path\"];\n\t\tif (glob) args.push(\"--glob\", glob);\n\t\targs.push(\"--\", pattern, cwd);\n\t\tconst child = spawn(rgPath, args, { stdio: [\"ignore\", \"pipe\", \"pipe\"] });\n\t\tconst rl = createInterface({ input: child.stdout });\n\t\tconst hits: GrepLineHit[] = [];\n\t\tlet stderr = \"\";\n\t\tlet killedDueToLimit = false;\n\t\tlet aborted = false;\n\n\t\tconst onAbort = () => {\n\t\t\taborted = true;\n\t\t\tif (!child.killed) child.kill();\n\t\t};\n\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\t\tchild.stderr?.on(\"data\", (chunk) => {\n\t\t\tstderr += chunk.toString();\n\t\t});\n\n\t\trl.on(\"line\", (line) => {\n\t\t\tif (!line.trim() || hits.length >= limit) return;\n\t\t\tlet event: any;\n\t\t\ttry {\n\t\t\t\tevent = JSON.parse(line);\n\t\t\t} catch {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (event.type !== \"match\") return;\n\t\t\tconst filePath = event.data?.path?.text;\n\t\t\tconst lineNumber = event.data?.line_number;\n\t\t\tif (filePath && typeof lineNumber === \"number\") {\n\t\t\t\thits.push({ rel: toRel(filePath), line: lineNumber, terms: termsOnLine(plan, event.data?.lines?.text) });\n\t\t\t}\n\t\t\tif (hits.length >= limit && !child.killed) {\n\t\t\t\tkilledDueToLimit = true;\n\t\t\t\tchild.kill();\n\t\t\t}\n\t\t});\n\n\t\tchild.on(\"error\", (error) => {\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\treject(new Error(`Failed to run ripgrep: ${error.message}`));\n\t\t});\n\t\tchild.on(\"close\", (code) => {\n\t\t\trl.close();\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\tif (aborted) {\n\t\t\t\treject(new Error(\"Operation aborted\"));\n\t\t\t\treturn;\n\t\t\t}\n\t\t\t// rg exits 1 on \"no matches\" — that is a valid empty result.\n\t\t\tif (!killedDueToLimit && code !== 0 && code !== 1) {\n\t\t\t\treject(new Error(stderr.trim() || `ripgrep exited with code ${code}`));\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tresolve(hits);\n\t\t});\n\t});\n}\n"]}
+{"version":3,"file":"lexical-retriever.js","sourceRoot":"","sources":["../../../src/core/search/lexical-retriever.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,eAAe,EAAE,MAAM,eAAe,CAAC;AAChD,OAAO,EAAE,KAAK,EAAE,MAAM,eAAe,CAAC;AACtC,OAAO,EAAE,YAAY,EAAE,MAAM,IAAI,CAAC;AAClC,OAAO,IAAI,MAAM,MAAM,CAAC;AACxB,OAAO,EAAE,UAAU,EAAE,MAAM,8BAA8B,CAAC;AAC1D,OAAO,EAAE,oBAAoB,EAAE,UAAU,EAAE,MAAM,2BAA2B,CAAC;AAG7E,4EAA4E;AAC5E,MAAM,SAAS,GAAG,CAAC,CAAC;AACpB,8CAA8C;AAC9C,MAAM,eAAe,GAAG,CAAC,CAAC;AAE1B,SAAS,YAAY,CAAC,KAAa,EAAU;IAC5C,OAAO,KAAK,CAAC,OAAO,CAAC,qBAAqB,EAAE,MAAM,CAAC,CAAC;AAAA,CACpD;AASD;;;;GAIG;AACH,MAAM,UAAU,qBAAqB,CAAC,KAAa,EAAgC;IAClF,MAAM,MAAM,GAAG,CAAC,GAAG,KAAK,CAAC,QAAQ,CAAC,sBAAsB,CAAC,CAAC;SACxD,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;SACvB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC;SAC3B,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;IACzC,IAAI,MAAM;QAAE,OAAO,EAAE,OAAO,EAAE,YAAY,CAAC,MAAM,CAAC,EAAE,KAAK,EAAE,CAAC,MAAM,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC;IAEpF,MAAM,MAAM,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,KAAK,CAAC,wBAAwB,CAAC,IAAI,EAAE,CAAC,CAAC;SACtE,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,IAAI,eAAe,CAAC;SAC1C,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC;SACzD,KAAK,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC;IACtB,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACzB,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC;QAC7B,OAAO,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,YAAY,CAAC,OAAO,CAAC,EAAE,KAAK,EAAE,CAAC,OAAO,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;IACjG,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,MAAM,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,EAAE,CAAC;AAAA,CAClG;AAED,0DAA0D;AAC1D,MAAM,UAAU,mBAAmB,CAAC,KAAa,EAAsB;IACtE,OAAO,qBAAqB,CAAC,KAAK,CAAC,EAAE,OAAO,CAAC;AAAA,CAC7C;AAED;8BAC8B;AAC9B,SAAS,WAAW,CAAC,IAAsB,EAAE,QAA4B,EAAY;IACpF,IAAI,CAAC,QAAQ;QAAE,OAAO,EAAE,CAAC;IACzB,MAAM,KAAK,GAAG,QAAQ,CAAC,WAAW,EAAE,CAAC;IACrC,OAAO,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;AAAA,CACnD;AAcD;;;;GAIG;AACH,SAAS,mBAAmB,CAAC,IAAwB,EAAsB;IAC1E,IAAI,CAAC,IAAI;QAAE,OAAO,SAAS,CAAC;IAC5B,yEAAyE;IACzE,6EAA6E;IAC7E,IAAI,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;QAC5E,OAAO,MAAM,IAAI,EAAE,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC;AAAA,CACZ;AAED,MAAM,CAAC,KAAK,UAAU,mBAAmB,CAAC,OAA0B,EAA0B;IAC7F,MAAM,EAAE,GAAG,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC;IAC7D,MAAM,IAAI,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;IAC1C,MAAM,IAAI,GAAG,qBAAqB,CAAC,KAAK,CAAC,CAAC;IAC1C,IAAI,CAAC,IAAI;QAAE,OAAO,EAAE,CAAC;IACrB,MAAM,EAAE,OAAO,EAAE,GAAG,IAAI,CAAC;IAEzB,MAAM,KAAK,GAAG,CAAC,QAAgB,EAAU,EAAE,CAAC;QAC3C,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAC;QACzC,OAAO,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,GAAG,CAAC,CAAC;IAAA,CAC3E,CAAC;IAEF,MAAM,MAAM,GAAG,oBAAoB,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,MAAM,UAAU,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC;IACjF,IAAI,CAAC,MAAM,EAAE,CAAC;QACb,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,GAAG,EAAE;YACpC,OAAO;YACP,WAAW,EAAE,IAAI;YACjB,UAAU,EAAE,IAAI;YAChB,KAAK;YACL,IAAI;YACJ,MAAM;YACN,QAAQ,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,CAAC,EAAE,OAAO,CAAC;SACzC,CAAC,CAAC;QACH,OAAO,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YACjC,GAAG,EAAE,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC;YACtB,IAAI,EAAE,CAAC,CAAC,UAAU;YAClB,KAAK,EAAE,WAAW,CAAC,IAAI,EAAE,CAAC,CAAC,QAAQ,CAAC;SACpC,CAAC,CAAC,CAAC;IACL,CAAC;IAED,OAAO,IAAI,OAAO,CAAgB,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE,CAAC;QACtD,sEAAsE;QACtE,kEAAkE;QAClE,oEAAkE;QAClE,iEAAiE;QACjE,mEAAmE;QACnE,MAAM,IAAI,GAAG;YACZ,QAAQ;YACR,eAAe;YACf,eAAe;YACf,UAAU;YACV,kBAAkB;YAClB,eAAe;YACf,QAAQ;YACR,MAAM;YACN,QAAQ;YACR,aAAa;SACb,CAAC;QACF,IAAI,IAAI;YAAE,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;QACpC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,EAAE,GAAG,CAAC,CAAC;QAC9B,MAAM,KAAK,GAAG,KAAK,CAAC,MAAM,EAAE,IAAI,EAAE,EAAE,KAAK,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC;QACzE,MAAM,EAAE,GAAG,eAAe,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,CAAC,CAAC;QACpD,MAAM,IAAI,GAAkB,EAAE,CAAC;QAC/B,IAAI,MAAM,GAAG,EAAE,CAAC;QAChB,IAAI,gBAAgB,GAAG,KAAK,CAAC;QAC7B,IAAI,OAAO,GAAG,KAAK,CAAC;QAEpB,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;YACrB,OAAO,GAAG,IAAI,CAAC;YACf,IAAI,CAAC,KAAK,CAAC,MAAM;gBAAE,KAAK,CAAC,IAAI,EAAE,CAAC;QAAA,CAChC,CAAC;QACF,MAAM,EAAE,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;QAC3D,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC;YACnC,MAAM,IAAI,KAAK,CAAC,QAAQ,EAAE,CAAC;QAAA,CAC3B,CAAC,CAAC;QAEH,EAAE,CAAC,EAAE,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC;YACvB,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,IAAI,CAAC,MAAM,IAAI,KAAK;gBAAE,OAAO;YACjD,IAAI,KAAU,CAAC;YACf,IAAI,CAAC;gBACJ,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;YAC1B,CAAC;YAAC,MAAM,CAAC;gBACR,OAAO;YACR,CAAC;YACD,IAAI,KAAK,CAAC,IAAI,KAAK,OAAO;gBAAE,OAAO;YACnC,MAAM,QAAQ,GAAG,KAAK,CAAC,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC;YACxC,MAAM,UAAU,GAAG,KAAK,CAAC,IAAI,EAAE,WAAW,CAAC;YAC3C,IAAI,QAAQ,IAAI,OAAO,UAAU,KAAK,QAAQ,EAAE,CAAC;gBAChD,IAAI,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,KAAK,CAAC,QAAQ,CAAC,EAAE,IAAI,EAAE,UAAU,EAAE,KAAK,EAAE,WAAW,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,EAAE,CAAC,CAAC;YAC1G,CAAC;YACD,IAAI,IAAI,CAAC,MAAM,IAAI,KAAK,IAAI,CAAC,KAAK,CAAC,MAAM,EAAE,CAAC;gBAC3C,gBAAgB,GAAG,IAAI,CAAC;gBACxB,KAAK,CAAC,IAAI,EAAE,CAAC;YACd,CAAC;QAAA,CACD,CAAC,CAAC;QAEH,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE,CAAC;YAC5B,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9C,MAAM,CAAC,IAAI,KAAK,CAAC,0BAA0B,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;QAAA,CAC7D,CAAC,CAAC;QACH,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,IAAI,EAAE,EAAE,CAAC;YAC3B,EAAE,CAAC,KAAK,EAAE,CAAC;YACX,MAAM,EAAE,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9C,IAAI,OAAO,EAAE,CAAC;gBACb,MAAM,CAAC,IAAI,KAAK,CAAC,mBAAmB,CAAC,CAAC,CAAC;gBACvC,OAAO;YACR,CAAC;YACD,+DAA6D;YAC7D,IAAI,CAAC,gBAAgB,IAAI,IAAI,KAAK,CAAC,IAAI,IAAI,KAAK,CAAC,EAAE,CAAC;gBACnD,MAAM,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,IAAI,EAAE,IAAI,4BAA4B,IAAI,EAAE,CAAC,CAAC,CAAC;gBACvE,OAAO;YACR,CAAC;YACD,OAAO,CAAC,IAAI,CAAC,CAAC;QAAA,CACd,CAAC,CAAC;IAAA,CACH,CAAC,CAAC;AAAA,CACH","sourcesContent":["/**\n * Internal lexical retriever for hybrid search.\n *\n * This is not the grep *tool* — it is the lexical recall backend: it turns a\n * natural query into a ripgrep pattern, streams matches, and returns bare\n * `rel:line` hits for the grep→chunk adapter. rg drives the fast path; the\n * pure-JS nativeGrep fallback keeps restricted environments working, same as\n * the grep tool.\n */\n\nimport { createInterface } from \"node:readline\";\nimport { spawn } from \"child_process\";\nimport { readFileSync } from \"fs\";\nimport path from \"path\";\nimport { ensureTool } from \"../../utils/tools-manager.js\";\nimport { isNativeSearchForced, nativeGrep } from \"../tools/native-search.js\";\nimport type { GrepLineHit } from \"./adapter.js\";\n\n/** Terms considered per query (longest first) when building the pattern. */\nconst MAX_TERMS = 4;\n/** Minimum token length worth matching on. */\nconst MIN_TERM_LENGTH = 3;\n\nfunction escapeRegExp(value: string): string {\n\treturn value.replace(/[.*+?^${}()|[\\]\\\\]/g, \"\\\\$&\");\n}\n\nexport interface LexicalQueryPlan {\n\t/** rg-ready regex pattern. */\n\tpattern: string;\n\t/** Raw (unescaped, lowercased) terms, for per-line term attribution. */\n\tterms: string[];\n}\n\n/**\n * Build the retrieval plan for a query: a quoted segment is searched\n * verbatim; otherwise the longest few identifier-ish tokens are OR-ed\n * together. Returns undefined when the query yields nothing searchable.\n */\nexport function buildLexicalQueryPlan(query: string): LexicalQueryPlan | undefined {\n\tconst quoted = [...query.matchAll(/[\"'`]([^\"'`]+)[\"'`]/g)]\n\t\t.map((m) => m[1].trim())\n\t\t.filter((s) => s.length > 0)\n\t\t.sort((a, b) => b.length - a.length)[0];\n\tif (quoted) return { pattern: escapeRegExp(quoted), terms: [quoted.toLowerCase()] };\n\n\tconst tokens = [...new Set(query.match(/[A-Za-z0-9_$][\\w$.-]*/g) ?? [])]\n\t\t.filter((t) => t.length >= MIN_TERM_LENGTH)\n\t\t.sort((a, b) => b.length - a.length || a.localeCompare(b))\n\t\t.slice(0, MAX_TERMS);\n\tif (tokens.length === 0) {\n\t\tconst trimmed = query.trim();\n\t\treturn trimmed ? { pattern: escapeRegExp(trimmed), terms: [trimmed.toLowerCase()] } : undefined;\n\t}\n\treturn { pattern: tokens.map(escapeRegExp).join(\"|\"), terms: tokens.map((t) => t.toLowerCase()) };\n}\n\n/** Pattern-only view of {@link buildLexicalQueryPlan}. */\nexport function buildLexicalPattern(query: string): string | undefined {\n\treturn buildLexicalQueryPlan(query)?.pattern;\n}\n\n/** Which plan terms appear on a matched line (retrieval is case-insensitive,\n *  so attribution is too). */\nfunction termsOnLine(plan: LexicalQueryPlan, lineText: string | undefined): string[] {\n\tif (!lineText) return [];\n\tconst lower = lineText.toLowerCase();\n\treturn plan.terms.filter((t) => lower.includes(t));\n}\n\n/**\n * Run lexical retrieval over `cwd`, returning up to `limit` line-hits in\n * output order with POSIX repo-relative paths.\n */\nexport interface RunLexicalOptions {\n\tcwd: string;\n\tquery: string;\n\tlimit: number;\n\tglob?: string;\n\tsignal?: AbortSignal;\n}\n\n/**\n * Run the lexical retriever for the search tool. Optional glob filter scopes\n * file paths (slashless matches basename anywhere, slash patterns match the\n * full repo-relative path).\n */\nfunction normalizeSearchGlob(glob: string | undefined): string | undefined {\n\tif (!glob) return undefined;\n\t// Match fd/rg semantics: a slash-containing glob is anchored anywhere in\n\t// the tree, so prepend \"**/\" unless it already starts with a slash or \"**/\".\n\tif (glob.includes(\"/\") && !glob.startsWith(\"/\") && !glob.startsWith(\"**/\")) {\n\t\treturn `**/${glob}`;\n\t}\n\treturn glob;\n}\n\nexport async function runLexicalRetriever(options: RunLexicalOptions): Promise<GrepLineHit[]> {\n\tconst { cwd, query, limit, glob: rawGlob, signal } = options;\n\tconst glob = normalizeSearchGlob(rawGlob);\n\tconst plan = buildLexicalQueryPlan(query);\n\tif (!plan) return [];\n\tconst { pattern } = plan;\n\n\tconst toRel = (filePath: string): string => {\n\t\tconst rel = path.relative(cwd, filePath);\n\t\treturn (rel && !rel.startsWith(\"..\") ? rel : filePath).replace(/\\\\/g, \"/\");\n\t};\n\n\tconst rgPath = isNativeSearchForced() ? undefined : await ensureTool(\"rg\", true);\n\tif (!rgPath) {\n\t\tconst result = await nativeGrep(cwd, {\n\t\t\tpattern,\n\t\t\tisDirectory: true,\n\t\t\tignoreCase: true,\n\t\t\tlimit,\n\t\t\tglob,\n\t\t\tsignal,\n\t\t\treadFile: (p) => readFileSync(p, \"utf-8\"),\n\t\t});\n\t\treturn result.matches.map((m) => ({\n\t\t\trel: toRel(m.filePath),\n\t\t\tline: m.lineNumber,\n\t\t\tterms: termsOnLine(plan, m.lineText),\n\t\t}));\n\t}\n\n\treturn new Promise<GrepLineHit[]>((resolve, reject) => {\n\t\t// --sort path forces a deterministic (single-threaded) walk: with the\n\t\t// match cap truncating the stream, a parallel walk would return a\n\t\t// different hit subset per run — \"same query, different context\".\n\t\t// `!.git` because --hidden would otherwise search .git contents;\n\t\t// --no-require-git so .gitignore is honored outside git repos too.\n\t\tconst args = [\n\t\t\t\"--json\",\n\t\t\t\"--line-number\",\n\t\t\t\"--color=never\",\n\t\t\t\"--hidden\",\n\t\t\t\"--no-require-git\",\n\t\t\t\"--ignore-case\",\n\t\t\t\"--sort\",\n\t\t\t\"path\",\n\t\t\t\"--glob\",\n\t\t\t\"!**/.git/**\",\n\t\t];\n\t\tif (glob) args.push(\"--glob\", glob);\n\t\targs.push(\"--\", pattern, cwd);\n\t\tconst child = spawn(rgPath, args, { stdio: [\"ignore\", \"pipe\", \"pipe\"] });\n\t\tconst rl = createInterface({ input: child.stdout });\n\t\tconst hits: GrepLineHit[] = [];\n\t\tlet stderr = \"\";\n\t\tlet killedDueToLimit = false;\n\t\tlet aborted = false;\n\n\t\tconst onAbort = () => {\n\t\t\taborted = true;\n\t\t\tif (!child.killed) child.kill();\n\t\t};\n\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\t\tchild.stderr?.on(\"data\", (chunk) => {\n\t\t\tstderr += chunk.toString();\n\t\t});\n\n\t\trl.on(\"line\", (line) => {\n\t\t\tif (!line.trim() || hits.length >= limit) return;\n\t\t\tlet event: any;\n\t\t\ttry {\n\t\t\t\tevent = JSON.parse(line);\n\t\t\t} catch {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (event.type !== \"match\") return;\n\t\t\tconst filePath = event.data?.path?.text;\n\t\t\tconst lineNumber = event.data?.line_number;\n\t\t\tif (filePath && typeof lineNumber === \"number\") {\n\t\t\t\thits.push({ rel: toRel(filePath), line: lineNumber, terms: termsOnLine(plan, event.data?.lines?.text) });\n\t\t\t}\n\t\t\tif (hits.length >= limit && !child.killed) {\n\t\t\t\tkilledDueToLimit = true;\n\t\t\t\tchild.kill();\n\t\t\t}\n\t\t});\n\n\t\tchild.on(\"error\", (error) => {\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\treject(new Error(`Failed to run ripgrep: ${error.message}`));\n\t\t});\n\t\tchild.on(\"close\", (code) => {\n\t\t\trl.close();\n\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\tif (aborted) {\n\t\t\t\treject(new Error(\"Operation aborted\"));\n\t\t\t\treturn;\n\t\t\t}\n\t\t\t// rg exits 1 on \"no matches\" — that is a valid empty result.\n\t\t\tif (!killedDueToLimit && code !== 0 && code !== 1) {\n\t\t\t\treject(new Error(stderr.trim() || `ripgrep exited with code ${code}`));\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tresolve(hits);\n\t\t});\n\t});\n}\n"]}

package/dist/core/tools/find.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"find.d.ts","sourceRoot":"","sources":["../../../src/core/tools/find.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,iCAAiC,CAAC;AAKjE,OAAO,EAAE,KAAK,MAAM,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAG5C,OAAO,KAAK,EAAE,cAAc,EAA2B,MAAM,wBAAwB,CAAC;AAMtF,OAAO,EAAiC,KAAK,gBAAgB,EAAgB,MAAM,eAAe,CAAC;AAEnG,QAAA,MAAM,UAAU;;;;;;;;EAqBd,CAAC;AAEH,MAAM,MAAM,aAAa,GAAG,MAAM,CAAC,OAAO,UAAU,CAAC,CAAC;AAKtD,MAAM,WAAW,eAAe;IAC/B,UAAU,CAAC,EAAE,gBAAgB,CAAC;IAC9B,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC5B;AAED;;;GAGG;AACH,MAAM,WAAW,cAAc;IAC9B,2BAA2B;IAC3B,MAAM,EAAE,CAAC,YAAY,EAAE,MAAM,KAAK,OAAO,CAAC,OAAO,CAAC,GAAG,OAAO,CAAC;IAC7D,yFAAyF;IACzF,IAAI,EAAE,CACL,QAAQ,EAAE,MAAM,EAAE,EAClB,GAAG,EAAE,MAAM,EACX,OAAO,EAAE;QAAE,MAAM,EAAE,MAAM,EAAE,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAA;KAAE,KACvD,OAAO,CAAC,MAAM,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC;CAClC;AAQD,MAAM,WAAW,eAAe;IAC/B,oEAAoE;IACpE,UAAU,CAAC,EAAE,cAAc,CAAC;CAC5B;AAoID,wBAAgB,wBAAwB,CACvC,GAAG,EAAE,MAAM,EACX,OAAO,CAAC,EAAE,eAAe,GACvB,cAAc,CAAC,OAAO,UAAU,EAAE,eAAe,GAAG,SAAS,CAAC,CAiPhE;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,SAAS,CAAC,OAAO,UAAU,CAAC,CAEnG","sourcesContent":["import { createInterface } from \"node:readline\";\nimport type { AgentTool } from \"@kolisachint/hoocode-agent-core\";\nimport { Text } from \"@kolisachint/hoocode-tui\";\nimport { type ChildProcess, spawn } from \"child_process\";\nimport { existsSync } from \"fs\";\nimport path from \"path\";\nimport { type Static, Type } from \"typebox\";\nimport { keyHint } from \"../../modes/interactive/components/keybinding-hints.js\";\nimport { ensureTool } from \"../../utils/tools-manager.js\";\nimport type { ToolDefinition, ToolRenderResultOptions } from \"../extensions/types.js\";\nimport { applyFdGlobPattern, relativizeFdLine, toPosixPath } from \"./fd-utils.js\";\nimport { isNativeSearchForced, nativeFind } from \"./native-search.js\";\nimport { resolveToCwd } from \"./path-utils.js\";\nimport { getTextOutput, invalidArgText, shortenPath, str } from \"./render-utils.js\";\nimport { wrapToolDefinition } from \"./tool-definition-wrapper.js\";\nimport { DEFAULT_MAX_BYTES, formatSize, type TruncationResult, truncateHead } from \"./truncate.js\";\n\nconst findSchema = Type.Object({\n\tpattern: Type.Union([Type.String(), Type.Array(Type.String())], {\n\t\tdescription:\n\t\t\t\"Glob pattern(s) to match files. Pass one pattern or an array for OR logic, e.g. '*.ts', 'src/**/*.spec.ts', or ['src/**/*.ts', 'test/**/*.ts'].\",\n\t}),\n\tpath: Type.Optional(Type.String({ description: \"Directory to search in (default: current directory)\" })),\n\texclude: Type.Optional(\n\t\tType.Union([Type.String(), Type.Array(Type.String())], {\n\t\t\tdescription: \"Additional exclusion glob(s), e.g. '**/*.test.ts' or ['**/dist/**', '**/build/**'].\",\n\t\t}),\n\t),\n\ttype: Type.Optional(\n\t\tType.Union([Type.Literal(\"f\"), Type.Literal(\"d\"), Type.Literal(\"l\")], {\n\t\t\tdescription: \"Filter by entry type: 'f' files, 'd' directories, 'l' symlinks (default: 'f').\",\n\t\t}),\n\t),\n\tdepth: Type.Optional(Type.Number({ description: \"Maximum directory depth to search.\" })),\n\tlimit: Type.Optional(Type.Number({ description: \"Maximum number of results (default: 1000).\" })),\n\tcompress: Type.Optional(\n\t\tType.Boolean({ description: \"Group files in the same directory to shorten output (default: false).\" }),\n\t),\n});\n\nexport type FindToolInput = Static<typeof findSchema>;\n\nconst DEFAULT_LIMIT = 1000;\nconst NO_RESULTS_MESSAGE = \"No files found matching pattern\";\n\nexport interface FindToolDetails {\n\ttruncation?: TruncationResult;\n\tresultLimitReached?: number;\n}\n\n/**\n * Pluggable operations for the find tool.\n * Override these to delegate file search to remote systems (for example SSH).\n */\nexport interface FindOperations {\n\t/** Check if path exists */\n\texists: (absolutePath: string) => Promise<boolean> | boolean;\n\t/** Find files matching one or more glob patterns. Returns relative or absolute paths. */\n\tglob: (\n\t\tpatterns: string[],\n\t\tcwd: string,\n\t\toptions: { ignore: string[]; limit: number; type?: string },\n\t) => Promise<string[]> | string[];\n}\n\nconst defaultFindOperations: FindOperations = {\n\texists: existsSync,\n\t// This is a placeholder. Actual fd execution happens in execute() when no custom glob is provided.\n\tglob: () => [],\n};\n\nexport interface FindToolOptions {\n\t/** Custom operations for find. Default: local filesystem plus fd */\n\toperations?: FindOperations;\n}\n\n/**\n * Compress a list of file paths by grouping files in the same directory.\n *\n * Compression rules:\n * - If a directory has >=3 entries all sharing the same extension: fold into `dir/{stem1,stem2,stem3}.ext`\n * - If >=3 entries with mixed extensions: sub-group by extension; fold sub-groups >=3\n * - If directory has >6 total files: summarize verbatim after sub-grouping\n * - 1-2 entries: emit verbatim\n */\nfunction compressPaths(paths: string[]): string {\n\tif (paths.length === 0) return \"\";\n\n\tconst sorted = [...paths].sort();\n\n\tconst dirMap = new Map<string, string[]>();\n\tfor (const p of sorted) {\n\t\tconst dir = path.dirname(p);\n\t\tconst base = path.basename(p);\n\t\tif (!dirMap.has(dir)) dirMap.set(dir, []);\n\t\tdirMap.get(dir)!.push(base);\n\t}\n\n\tconst result: string[] = [];\n\n\tfor (const [dir, files] of dirMap) {\n\t\tif (files.length <= 2) {\n\t\t\tfor (const f of files) result.push(dir === \".\" ? f : `${dir}/${f}`);\n\t\t\tcontinue;\n\t\t}\n\n\t\tconst extMap = new Map<string, string[]>();\n\t\tfor (const f of files) {\n\t\t\tconst ext = path.extname(f);\n\t\t\tconst stem = path.basename(f, ext);\n\t\t\tif (!extMap.has(ext)) extMap.set(ext, []);\n\t\t\textMap.get(ext)!.push(stem);\n\t\t}\n\n\t\tif (extMap.size === 1 && files.length >= 3) {\n\t\t\tconst [ext, stems] = extMap.entries().next().value!;\n\t\t\tconst displayDir = dir === \".\" ? \"\" : `${dir}/`;\n\t\t\tresult.push(`${displayDir}{${stems.join(\",\")}}${ext}`);\n\t\t\tcontinue;\n\t\t}\n\n\t\tlet anyFolded = false;\n\t\tconst subResults: string[] = [];\n\t\tfor (const [ext, stems] of extMap) {\n\t\t\tif (stems.length >= 3) {\n\t\t\t\tconst displayDir = dir === \".\" ? \"\" : `${dir}/`;\n\t\t\t\tsubResults.push(`${displayDir}{${stems.join(\",\")}}${ext}`);\n\t\t\t\tanyFolded = true;\n\t\t\t} else {\n\t\t\t\tfor (const stem of stems) subResults.push(dir === \".\" ? `${stem}${ext}` : `${dir}/${stem}${ext}`);\n\t\t\t}\n\t\t}\n\n\t\tif (anyFolded || files.length > 6) {\n\t\t\tresult.push(...subResults);\n\t\t} else {\n\t\t\tfor (const f of files) result.push(dir === \".\" ? f : `${dir}/${f}`);\n\t\t}\n\t}\n\n\treturn result.join(\"\\n\");\n}\n\nfunction formatFindCall(\n\targs: { pattern?: string | string[]; path?: string; type?: string; depth?: number; limit?: number } | undefined,\n\ttheme: typeof import(\"../../modes/interactive/theme/theme.js\").theme,\n): string {\n\tconst patterns = args?.pattern;\n\tconst patternStr = Array.isArray(patterns) ? patterns.join(\", \") : str(patterns);\n\tconst rawPath = str(args?.path);\n\tconst displayPath = rawPath !== null ? shortenPath(rawPath || \".\") : null;\n\tconst type = args?.type;\n\tconst depth = args?.depth;\n\tconst limit = args?.limit;\n\tconst invalidArg = invalidArgText(theme);\n\tlet text =\n\t\ttheme.fg(\"toolTitle\", theme.bold(\"find\")) +\n\t\t\" \" +\n\t\t(patternStr === null || patternStr === \"\" ? invalidArg : theme.fg(\"accent\", patternStr)) +\n\t\ttheme.fg(\"toolOutput\", ` in ${displayPath === null ? invalidArg : displayPath}`);\n\tif (type && type !== \"f\") {\n\t\tconst typeLabel = type === \"d\" ? \"dirs\" : type === \"l\" ? \"symlinks\" : type;\n\t\ttext += theme.fg(\"toolOutput\", ` (${typeLabel})`);\n\t}\n\tif (depth !== undefined) {\n\t\ttext += theme.fg(\"toolOutput\", ` (depth ${depth})`);\n\t}\n\tif (limit !== undefined) {\n\t\ttext += theme.fg(\"toolOutput\", ` (limit ${limit})`);\n\t}\n\treturn text;\n}\n\nfunction formatFindResult(\n\tresult: {\n\t\tcontent: Array<{ type: string; text?: string; data?: string; mimeType?: string }>;\n\t\tdetails?: FindToolDetails;\n\t},\n\toptions: ToolRenderResultOptions,\n\ttheme: typeof import(\"../../modes/interactive/theme/theme.js\").theme,\n\tshowImages: boolean,\n): string {\n\tconst output = getTextOutput(result, showImages).trim();\n\tlet text = \"\";\n\tif (output) {\n\t\tconst lines = output.split(\"\\n\");\n\t\tconst maxLines = options.expanded ? lines.length : 20;\n\t\tconst displayLines = lines.slice(0, maxLines);\n\t\tconst remaining = lines.length - maxLines;\n\t\ttext += `\\n${displayLines.map((line) => theme.fg(\"toolOutput\", line)).join(\"\\n\")}`;\n\t\tif (remaining > 0) {\n\t\t\ttext += `${theme.fg(\"muted\", `\\n... (${remaining} more lines,`)} ${keyHint(\"app.tools.expand\", \"to expand\")})`;\n\t\t}\n\t}\n\n\tconst resultLimit = result.details?.resultLimitReached;\n\tconst truncation = result.details?.truncation;\n\tif (resultLimit || truncation?.truncated) {\n\t\tconst warnings: string[] = [];\n\t\tif (resultLimit) warnings.push(`${resultLimit} results limit`);\n\t\tif (truncation?.truncated) warnings.push(`${formatSize(truncation.maxBytes ?? DEFAULT_MAX_BYTES)} limit`);\n\t\ttext += `\\n${theme.fg(\"warning\", `[Truncated: ${warnings.join(\", \")}]`)}`;\n\t}\n\treturn text;\n}\n\nexport function createFindToolDefinition(\n\tcwd: string,\n\toptions?: FindToolOptions,\n): ToolDefinition<typeof findSchema, FindToolDetails | undefined> {\n\tconst customOps = options?.operations;\n\treturn {\n\t\tname: \"find\",\n\t\tlabel: \"find\",\n\t\tdescription: `Search for files by one or more glob patterns (OR logic across an array). Optionally filter by entry type (files/dirs/symlinks), directory depth, and extra exclusions. Returns matching paths relative to the search directory. Respects .gitignore. Output is truncated to ${DEFAULT_LIMIT} results or ${DEFAULT_MAX_BYTES / 1024}KB (whichever is hit first).`,\n\t\tpromptSnippet: \"Find files by glob pattern (supports multiple patterns, respects .gitignore)\",\n\t\tparameters: findSchema,\n\t\tasync execute(\n\t\t\t_toolCallId,\n\t\t\t{\n\t\t\t\tpattern: rawPattern,\n\t\t\t\tpath: searchDir,\n\t\t\t\texclude: rawExclude,\n\t\t\t\ttype,\n\t\t\t\tdepth,\n\t\t\t\tlimit,\n\t\t\t\tcompress = false,\n\t\t\t}: {\n\t\t\t\tpattern: string | string[];\n\t\t\t\tpath?: string;\n\t\t\t\texclude?: string | string[];\n\t\t\t\ttype?: \"f\" | \"d\" | \"l\";\n\t\t\t\tdepth?: number;\n\t\t\t\tlimit?: number;\n\t\t\t\tcompress?: boolean;\n\t\t\t},\n\t\t\tsignal?: AbortSignal,\n\t\t\t_onUpdate?,\n\t\t\t_ctx?,\n\t\t) {\n\t\t\treturn new Promise((resolve, reject) => {\n\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\treject(new Error(\"Operation aborted\"));\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\tlet settled = false;\n\t\t\t\tlet stopChildren: (() => void) | undefined;\n\t\t\t\tconst settle = (fn: () => void) => {\n\t\t\t\t\tif (settled) return;\n\t\t\t\t\tsettled = true;\n\t\t\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\t\t\tstopChildren = undefined;\n\t\t\t\t\tfn();\n\t\t\t\t};\n\t\t\t\tconst onAbort = () => {\n\t\t\t\t\tstopChildren?.();\n\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t};\n\t\t\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\n\t\t\t\t(async () => {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tconst searchPath = resolveToCwd(searchDir || \".\", cwd);\n\t\t\t\t\t\tconst effectiveLimit = limit ?? DEFAULT_LIMIT;\n\t\t\t\t\t\tconst ops = customOps ?? defaultFindOperations;\n\n\t\t\t\t\t\tconst patterns = Array.isArray(rawPattern) ? rawPattern : [rawPattern];\n\t\t\t\t\t\tconst excludePatterns = rawExclude ? (Array.isArray(rawExclude) ? rawExclude : [rawExclude]) : [];\n\t\t\t\t\t\t// Always exclude node_modules and .git, plus any caller exclusions.\n\t\t\t\t\t\tconst allIgnore = [\"**/node_modules/**\", \"**/.git/**\", ...excludePatterns];\n\t\t\t\t\t\tconst typeFilter = type === \"d\" ? \"d\" : type === \"l\" ? \"l\" : \"f\";\n\n\t\t\t\t\t\tconst emit = (relativized: string[]) => {\n\t\t\t\t\t\t\tif (relativized.length === 0) {\n\t\t\t\t\t\t\t\tsettle(() =>\n\t\t\t\t\t\t\t\t\tresolve({ content: [{ type: \"text\", text: NO_RESULTS_MESSAGE }], details: undefined }),\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tconst unique = [...new Set(relativized)].sort();\n\t\t\t\t\t\t\tconst resultLimitReached = unique.length >= effectiveLimit;\n\t\t\t\t\t\t\tconst truncated = unique.slice(0, effectiveLimit);\n\t\t\t\t\t\t\tconst rawOutput = compress ? compressPaths(truncated) : truncated.join(\"\\n\");\n\t\t\t\t\t\t\tconst truncation = truncateHead(rawOutput, { maxLines: Number.MAX_SAFE_INTEGER });\n\t\t\t\t\t\t\tlet resultOutput = truncation.content;\n\t\t\t\t\t\t\tconst details: FindToolDetails = {};\n\t\t\t\t\t\t\tconst notices: string[] = [];\n\t\t\t\t\t\t\tif (resultLimitReached) {\n\t\t\t\t\t\t\t\tnotices.push(\n\t\t\t\t\t\t\t\t\t`${effectiveLimit} results limit reached. Use limit=${effectiveLimit * 2} for more, or refine pattern`,\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\tdetails.resultLimitReached = effectiveLimit;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (truncation.truncated) {\n\t\t\t\t\t\t\t\tnotices.push(`${formatSize(DEFAULT_MAX_BYTES)} limit reached`);\n\t\t\t\t\t\t\t\tdetails.truncation = truncation;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (notices.length > 0) resultOutput += `\\n\\n[${notices.join(\". \")}]`;\n\t\t\t\t\t\t\tsettle(() =>\n\t\t\t\t\t\t\t\tresolve({\n\t\t\t\t\t\t\t\t\tcontent: [{ type: \"text\", text: resultOutput }],\n\t\t\t\t\t\t\t\t\tdetails: Object.keys(details).length > 0 ? details : undefined,\n\t\t\t\t\t\t\t\t}),\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t};\n\n\t\t\t\t\t\t// If custom operations provide glob(), use that instead of fd.\n\t\t\t\t\t\tif (customOps?.glob) {\n\t\t\t\t\t\t\tif (!(await ops.exists(searchPath))) {\n\t\t\t\t\t\t\t\tsettle(() => reject(new Error(`Path not found: ${searchPath}`)));\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tconst results = await ops.glob(patterns, searchPath, {\n\t\t\t\t\t\t\t\tignore: allIgnore,\n\t\t\t\t\t\t\t\tlimit: effectiveLimit,\n\t\t\t\t\t\t\t\ttype: typeFilter,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\temit(\n\t\t\t\t\t\t\t\tresults.map((p) => {\n\t\t\t\t\t\t\t\t\tif (p.startsWith(searchPath)) return toPosixPath(p.slice(searchPath.length + 1));\n\t\t\t\t\t\t\t\t\treturn toPosixPath(path.relative(searchPath, p));\n\t\t\t\t\t\t\t\t}),\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// Default implementation uses fd, with a pure-JS fallback when fd is\n\t\t\t\t\t\t// unavailable (restricted environments) or explicitly forced.\n\t\t\t\t\t\tconst fdPath = isNativeSearchForced() ? undefined : await ensureTool(\"fd\", true);\n\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\t\t\t\t\t\tif (!fdPath) {\n\t\t\t\t\t\t\tif (!(await ops.exists(searchPath))) {\n\t\t\t\t\t\t\t\tsettle(() => reject(new Error(`Path not found: ${searchPath}`)));\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tconst nativeResults = nativeFind(searchPath, {\n\t\t\t\t\t\t\t\tpatterns,\n\t\t\t\t\t\t\t\ttype: typeFilter,\n\t\t\t\t\t\t\t\texcludeGlobs: allIgnore,\n\t\t\t\t\t\t\t\tmaxDepth: depth,\n\t\t\t\t\t\t\t\t// find always excludes node_modules/.git; the exclude globs above\n\t\t\t\t\t\t\t\t// cover them too, but skipping the dirs avoids descending them.\n\t\t\t\t\t\t\t\talwaysSkipDirs: new Set([\".git\", \"node_modules\"]),\n\t\t\t\t\t\t\t\tsignal,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\temit(nativeResults);\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tconst children: ChildProcess[] = [];\n\t\t\t\t\t\tstopChildren = () => {\n\t\t\t\t\t\t\tfor (const child of children) if (!child.killed) child.kill();\n\t\t\t\t\t\t};\n\n\t\t\t\t\t\t// Run each pattern with fd. fd errors (e.g. an invalid glob) surface as a\n\t\t\t\t\t\t// rejection when the pattern produced no output, matching the shell's\n\t\t\t\t\t\t// behavior; a non-zero exit that still produced matches is tolerated.\n\t\t\t\t\t\tconst runPattern = (pattern: string): Promise<string[]> =>\n\t\t\t\t\t\t\tnew Promise<string[]>((res, rej) => {\n\t\t\t\t\t\t\t\tconst args: string[] = [\n\t\t\t\t\t\t\t\t\t\"--glob\",\n\t\t\t\t\t\t\t\t\t\"--color=never\",\n\t\t\t\t\t\t\t\t\t\"--hidden\",\n\t\t\t\t\t\t\t\t\t\"--no-require-git\",\n\t\t\t\t\t\t\t\t\t\"--type\",\n\t\t\t\t\t\t\t\t\ttypeFilter,\n\t\t\t\t\t\t\t\t\t\"--max-results\",\n\t\t\t\t\t\t\t\t\tString(effectiveLimit),\n\t\t\t\t\t\t\t\t];\n\t\t\t\t\t\t\t\tfor (const excl of allIgnore) args.push(\"--exclude\", excl);\n\t\t\t\t\t\t\t\tif (depth !== undefined) args.push(\"--max-depth\", String(depth));\n\t\t\t\t\t\t\t\tconst effectivePattern = applyFdGlobPattern(args, pattern);\n\t\t\t\t\t\t\t\targs.push(\"--\", effectivePattern, searchPath);\n\n\t\t\t\t\t\t\t\tconst child = spawn(fdPath, args, { stdio: [\"ignore\", \"pipe\", \"pipe\"] });\n\t\t\t\t\t\t\t\tchildren.push(child);\n\t\t\t\t\t\t\t\tconst rl = createInterface({ input: child.stdout! });\n\t\t\t\t\t\t\t\tlet stderr = \"\";\n\t\t\t\t\t\t\t\tconst lines: string[] = [];\n\t\t\t\t\t\t\t\tchild.stderr?.on(\"data\", (chunk) => {\n\t\t\t\t\t\t\t\t\tstderr += chunk.toString();\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\trl.on(\"line\", (line) => lines.push(line));\n\t\t\t\t\t\t\t\tchild.on(\"error\", (error) => {\n\t\t\t\t\t\t\t\t\trl.close();\n\t\t\t\t\t\t\t\t\trej(new Error(`Failed to run fd: ${error.message}`));\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\tchild.on(\"close\", (code) => {\n\t\t\t\t\t\t\t\t\trl.close();\n\t\t\t\t\t\t\t\t\tif (code !== 0 && lines.length === 0) {\n\t\t\t\t\t\t\t\t\t\trej(new Error(stderr.trim() || `fd exited with code ${code}`));\n\t\t\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\tres(lines);\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t});\n\n\t\t\t\t\t\tconst patternResults = await Promise.all(patterns.map(runPattern));\n\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tconst relativized: string[] = [];\n\t\t\t\t\t\tfor (const results of patternResults) {\n\t\t\t\t\t\t\tfor (const rawLine of results) {\n\t\t\t\t\t\t\t\tconst line = rawLine.replace(/\\r$/, \"\").trim();\n\t\t\t\t\t\t\t\tif (!line) continue;\n\t\t\t\t\t\t\t\trelativized.push(relativizeFdLine(line, searchPath));\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t\temit(relativized);\n\t\t\t\t\t} catch (e) {\n\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\t\t\t\t\t\tconst error = e instanceof Error ? e : new Error(String(e));\n\t\t\t\t\t\tsettle(() => reject(error));\n\t\t\t\t\t}\n\t\t\t\t})();\n\t\t\t});\n\t\t},\n\t\trenderCall(args, theme, context) {\n\t\t\tconst text = (context.lastComponent as Text | undefined) ?? new Text(\"\", 0, 0);\n\t\t\ttext.setText(formatFindCall(args, theme));\n\t\t\treturn text;\n\t\t},\n\t\trenderResult(result, options, theme, context) {\n\t\t\tconst text = (context.lastComponent as Text | undefined) ?? new Text(\"\", 0, 0);\n\t\t\ttext.setText(formatFindResult(result as any, options, theme, context.showImages));\n\t\t\treturn text;\n\t\t},\n\t};\n}\n\nexport function createFindTool(cwd: string, options?: FindToolOptions): AgentTool<typeof findSchema> {\n\treturn wrapToolDefinition(createFindToolDefinition(cwd, options));\n}\n"]}
+{"version":3,"file":"find.d.ts","sourceRoot":"","sources":["../../../src/core/tools/find.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,iCAAiC,CAAC;AAKjE,OAAO,EAAE,KAAK,MAAM,EAAE,IAAI,EAAE,MAAM,SAAS,CAAC;AAG5C,OAAO,KAAK,EAAE,cAAc,EAA2B,MAAM,wBAAwB,CAAC;AAMtF,OAAO,EAAiC,KAAK,gBAAgB,EAAgB,MAAM,eAAe,CAAC;AAEnG,QAAA,MAAM,UAAU;;;;;;;;EAqBd,CAAC;AAEH,MAAM,MAAM,aAAa,GAAG,MAAM,CAAC,OAAO,UAAU,CAAC,CAAC;AAKtD,MAAM,WAAW,eAAe;IAC/B,UAAU,CAAC,EAAE,gBAAgB,CAAC;IAC9B,kBAAkB,CAAC,EAAE,MAAM,CAAC;CAC5B;AAED;;;GAGG;AACH,MAAM,WAAW,cAAc;IAC9B,2BAA2B;IAC3B,MAAM,EAAE,CAAC,YAAY,EAAE,MAAM,KAAK,OAAO,CAAC,OAAO,CAAC,GAAG,OAAO,CAAC;IAC7D,yFAAyF;IACzF,IAAI,EAAE,CACL,QAAQ,EAAE,MAAM,EAAE,EAClB,GAAG,EAAE,MAAM,EACX,OAAO,EAAE;QAAE,MAAM,EAAE,MAAM,EAAE,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE,MAAM,CAAA;KAAE,KACvD,OAAO,CAAC,MAAM,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC;CAClC;AAQD,MAAM,WAAW,eAAe;IAC/B,oEAAoE;IACpE,UAAU,CAAC,EAAE,cAAc,CAAC;CAC5B;AAoID,wBAAgB,wBAAwB,CACvC,GAAG,EAAE,MAAM,EACX,OAAO,CAAC,EAAE,eAAe,GACvB,cAAc,CAAC,OAAO,UAAU,EAAE,eAAe,GAAG,SAAS,CAAC,CAmPhE;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,SAAS,CAAC,OAAO,UAAU,CAAC,CAEnG","sourcesContent":["import { createInterface } from \"node:readline\";\nimport type { AgentTool } from \"@kolisachint/hoocode-agent-core\";\nimport { Text } from \"@kolisachint/hoocode-tui\";\nimport { type ChildProcess, spawn } from \"child_process\";\nimport { existsSync } from \"fs\";\nimport path from \"path\";\nimport { type Static, Type } from \"typebox\";\nimport { keyHint } from \"../../modes/interactive/components/keybinding-hints.js\";\nimport { ensureTool } from \"../../utils/tools-manager.js\";\nimport type { ToolDefinition, ToolRenderResultOptions } from \"../extensions/types.js\";\nimport { applyFdGlobPattern, relativizeFdLine, toPosixPath } from \"./fd-utils.js\";\nimport { isNativeSearchForced, nativeFind } from \"./native-search.js\";\nimport { resolveToCwd } from \"./path-utils.js\";\nimport { getTextOutput, invalidArgText, shortenPath, str } from \"./render-utils.js\";\nimport { wrapToolDefinition } from \"./tool-definition-wrapper.js\";\nimport { DEFAULT_MAX_BYTES, formatSize, type TruncationResult, truncateHead } from \"./truncate.js\";\n\nconst findSchema = Type.Object({\n\tpattern: Type.Union([Type.String(), Type.Array(Type.String())], {\n\t\tdescription:\n\t\t\t\"Glob pattern(s) to match files. Pass one pattern or an array for OR logic, e.g. '*.ts', 'src/**/*.spec.ts', or ['src/**/*.ts', 'test/**/*.ts'].\",\n\t}),\n\tpath: Type.Optional(Type.String({ description: \"Directory to search in (default: current directory)\" })),\n\texclude: Type.Optional(\n\t\tType.Union([Type.String(), Type.Array(Type.String())], {\n\t\t\tdescription: \"Additional exclusion glob(s), e.g. '**/*.test.ts' or ['**/dist/**', '**/build/**'].\",\n\t\t}),\n\t),\n\ttype: Type.Optional(\n\t\tType.Union([Type.Literal(\"f\"), Type.Literal(\"d\"), Type.Literal(\"l\")], {\n\t\t\tdescription: \"Filter by entry type: 'f' files, 'd' directories, 'l' symlinks (default: 'f').\",\n\t\t}),\n\t),\n\tdepth: Type.Optional(Type.Number({ description: \"Maximum directory depth to search.\" })),\n\tlimit: Type.Optional(Type.Number({ description: \"Maximum number of results (default: 1000).\" })),\n\tcompress: Type.Optional(\n\t\tType.Boolean({ description: \"Group files in the same directory to shorten output (default: false).\" }),\n\t),\n});\n\nexport type FindToolInput = Static<typeof findSchema>;\n\nconst DEFAULT_LIMIT = 1000;\nconst NO_RESULTS_MESSAGE = \"No files found matching pattern\";\n\nexport interface FindToolDetails {\n\ttruncation?: TruncationResult;\n\tresultLimitReached?: number;\n}\n\n/**\n * Pluggable operations for the find tool.\n * Override these to delegate file search to remote systems (for example SSH).\n */\nexport interface FindOperations {\n\t/** Check if path exists */\n\texists: (absolutePath: string) => Promise<boolean> | boolean;\n\t/** Find files matching one or more glob patterns. Returns relative or absolute paths. */\n\tglob: (\n\t\tpatterns: string[],\n\t\tcwd: string,\n\t\toptions: { ignore: string[]; limit: number; type?: string },\n\t) => Promise<string[]> | string[];\n}\n\nconst defaultFindOperations: FindOperations = {\n\texists: existsSync,\n\t// This is a placeholder. Actual fd execution happens in execute() when no custom glob is provided.\n\tglob: () => [],\n};\n\nexport interface FindToolOptions {\n\t/** Custom operations for find. Default: local filesystem plus fd */\n\toperations?: FindOperations;\n}\n\n/**\n * Compress a list of file paths by grouping files in the same directory.\n *\n * Compression rules:\n * - If a directory has >=3 entries all sharing the same extension: fold into `dir/{stem1,stem2,stem3}.ext`\n * - If >=3 entries with mixed extensions: sub-group by extension; fold sub-groups >=3\n * - If directory has >6 total files: summarize verbatim after sub-grouping\n * - 1-2 entries: emit verbatim\n */\nfunction compressPaths(paths: string[]): string {\n\tif (paths.length === 0) return \"\";\n\n\tconst sorted = [...paths].sort();\n\n\tconst dirMap = new Map<string, string[]>();\n\tfor (const p of sorted) {\n\t\tconst dir = path.dirname(p);\n\t\tconst base = path.basename(p);\n\t\tif (!dirMap.has(dir)) dirMap.set(dir, []);\n\t\tdirMap.get(dir)!.push(base);\n\t}\n\n\tconst result: string[] = [];\n\n\tfor (const [dir, files] of dirMap) {\n\t\tif (files.length <= 2) {\n\t\t\tfor (const f of files) result.push(dir === \".\" ? f : `${dir}/${f}`);\n\t\t\tcontinue;\n\t\t}\n\n\t\tconst extMap = new Map<string, string[]>();\n\t\tfor (const f of files) {\n\t\t\tconst ext = path.extname(f);\n\t\t\tconst stem = path.basename(f, ext);\n\t\t\tif (!extMap.has(ext)) extMap.set(ext, []);\n\t\t\textMap.get(ext)!.push(stem);\n\t\t}\n\n\t\tif (extMap.size === 1 && files.length >= 3) {\n\t\t\tconst [ext, stems] = extMap.entries().next().value!;\n\t\t\tconst displayDir = dir === \".\" ? \"\" : `${dir}/`;\n\t\t\tresult.push(`${displayDir}{${stems.join(\",\")}}${ext}`);\n\t\t\tcontinue;\n\t\t}\n\n\t\tlet anyFolded = false;\n\t\tconst subResults: string[] = [];\n\t\tfor (const [ext, stems] of extMap) {\n\t\t\tif (stems.length >= 3) {\n\t\t\t\tconst displayDir = dir === \".\" ? \"\" : `${dir}/`;\n\t\t\t\tsubResults.push(`${displayDir}{${stems.join(\",\")}}${ext}`);\n\t\t\t\tanyFolded = true;\n\t\t\t} else {\n\t\t\t\tfor (const stem of stems) subResults.push(dir === \".\" ? `${stem}${ext}` : `${dir}/${stem}${ext}`);\n\t\t\t}\n\t\t}\n\n\t\tif (anyFolded || files.length > 6) {\n\t\t\tresult.push(...subResults);\n\t\t} else {\n\t\t\tfor (const f of files) result.push(dir === \".\" ? f : `${dir}/${f}`);\n\t\t}\n\t}\n\n\treturn result.join(\"\\n\");\n}\n\nfunction formatFindCall(\n\targs: { pattern?: string | string[]; path?: string; type?: string; depth?: number; limit?: number } | undefined,\n\ttheme: typeof import(\"../../modes/interactive/theme/theme.js\").theme,\n): string {\n\tconst patterns = args?.pattern;\n\tconst patternStr = Array.isArray(patterns) ? patterns.join(\", \") : str(patterns);\n\tconst rawPath = str(args?.path);\n\tconst displayPath = rawPath !== null ? shortenPath(rawPath || \".\") : null;\n\tconst type = args?.type;\n\tconst depth = args?.depth;\n\tconst limit = args?.limit;\n\tconst invalidArg = invalidArgText(theme);\n\tlet text =\n\t\ttheme.fg(\"toolTitle\", theme.bold(\"find\")) +\n\t\t\" \" +\n\t\t(patternStr === null || patternStr === \"\" ? invalidArg : theme.fg(\"accent\", patternStr)) +\n\t\ttheme.fg(\"toolOutput\", ` in ${displayPath === null ? invalidArg : displayPath}`);\n\tif (type && type !== \"f\") {\n\t\tconst typeLabel = type === \"d\" ? \"dirs\" : type === \"l\" ? \"symlinks\" : type;\n\t\ttext += theme.fg(\"toolOutput\", ` (${typeLabel})`);\n\t}\n\tif (depth !== undefined) {\n\t\ttext += theme.fg(\"toolOutput\", ` (depth ${depth})`);\n\t}\n\tif (limit !== undefined) {\n\t\ttext += theme.fg(\"toolOutput\", ` (limit ${limit})`);\n\t}\n\treturn text;\n}\n\nfunction formatFindResult(\n\tresult: {\n\t\tcontent: Array<{ type: string; text?: string; data?: string; mimeType?: string }>;\n\t\tdetails?: FindToolDetails;\n\t},\n\toptions: ToolRenderResultOptions,\n\ttheme: typeof import(\"../../modes/interactive/theme/theme.js\").theme,\n\tshowImages: boolean,\n): string {\n\tconst output = getTextOutput(result, showImages).trim();\n\tlet text = \"\";\n\tif (output) {\n\t\tconst lines = output.split(\"\\n\");\n\t\tconst maxLines = options.expanded ? lines.length : 20;\n\t\tconst displayLines = lines.slice(0, maxLines);\n\t\tconst remaining = lines.length - maxLines;\n\t\ttext += `\\n${displayLines.map((line) => theme.fg(\"toolOutput\", line)).join(\"\\n\")}`;\n\t\tif (remaining > 0) {\n\t\t\ttext += `${theme.fg(\"muted\", `\\n... (${remaining} more lines,`)} ${keyHint(\"app.tools.expand\", \"to expand\")})`;\n\t\t}\n\t}\n\n\tconst resultLimit = result.details?.resultLimitReached;\n\tconst truncation = result.details?.truncation;\n\tif (resultLimit || truncation?.truncated) {\n\t\tconst warnings: string[] = [];\n\t\tif (resultLimit) warnings.push(`${resultLimit} results limit`);\n\t\tif (truncation?.truncated) warnings.push(`${formatSize(truncation.maxBytes ?? DEFAULT_MAX_BYTES)} limit`);\n\t\ttext += `\\n${theme.fg(\"warning\", `[Truncated: ${warnings.join(\", \")}]`)}`;\n\t}\n\treturn text;\n}\n\nexport function createFindToolDefinition(\n\tcwd: string,\n\toptions?: FindToolOptions,\n): ToolDefinition<typeof findSchema, FindToolDetails | undefined> {\n\tconst customOps = options?.operations;\n\treturn {\n\t\tname: \"find\",\n\t\tlabel: \"find\",\n\t\tdescription: `Search for files by one or more glob patterns (OR logic across an array). Optionally filter by entry type (files/dirs/symlinks), directory depth, and extra exclusions. Returns matching paths relative to the search directory. Respects .gitignore. Output is truncated to ${DEFAULT_LIMIT} results or ${DEFAULT_MAX_BYTES / 1024}KB (whichever is hit first).`,\n\t\tpromptSnippet: \"Find files by glob pattern (supports multiple patterns, respects .gitignore)\",\n\t\tparameters: findSchema,\n\t\tasync execute(\n\t\t\t_toolCallId,\n\t\t\t{\n\t\t\t\tpattern: rawPattern,\n\t\t\t\tpath: searchDir,\n\t\t\t\texclude: rawExclude,\n\t\t\t\ttype,\n\t\t\t\tdepth,\n\t\t\t\tlimit,\n\t\t\t\tcompress = false,\n\t\t\t}: {\n\t\t\t\tpattern: string | string[];\n\t\t\t\tpath?: string;\n\t\t\t\texclude?: string | string[];\n\t\t\t\ttype?: \"f\" | \"d\" | \"l\";\n\t\t\t\tdepth?: number;\n\t\t\t\tlimit?: number;\n\t\t\t\tcompress?: boolean;\n\t\t\t},\n\t\t\tsignal?: AbortSignal,\n\t\t\t_onUpdate?,\n\t\t\t_ctx?,\n\t\t) {\n\t\t\treturn new Promise((resolve, reject) => {\n\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\treject(new Error(\"Operation aborted\"));\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\tlet settled = false;\n\t\t\t\tlet stopChildren: (() => void) | undefined;\n\t\t\t\tconst settle = (fn: () => void) => {\n\t\t\t\t\tif (settled) return;\n\t\t\t\t\tsettled = true;\n\t\t\t\t\tsignal?.removeEventListener(\"abort\", onAbort);\n\t\t\t\t\tstopChildren = undefined;\n\t\t\t\t\tfn();\n\t\t\t\t};\n\t\t\t\tconst onAbort = () => {\n\t\t\t\t\tstopChildren?.();\n\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t};\n\t\t\t\tsignal?.addEventListener(\"abort\", onAbort, { once: true });\n\n\t\t\t\t(async () => {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tconst searchPath = resolveToCwd(searchDir || \".\", cwd);\n\t\t\t\t\t\tconst effectiveLimit = Math.max(1, limit ?? DEFAULT_LIMIT);\n\t\t\t\t\t\tconst ops = customOps ?? defaultFindOperations;\n\n\t\t\t\t\t\tconst patterns = Array.isArray(rawPattern) ? rawPattern : [rawPattern];\n\t\t\t\t\t\tconst excludePatterns = rawExclude ? (Array.isArray(rawExclude) ? rawExclude : [rawExclude]) : [];\n\t\t\t\t\t\t// Always exclude node_modules and .git, plus any caller exclusions.\n\t\t\t\t\t\tconst allIgnore = [\"**/node_modules/**\", \"**/.git/**\", ...excludePatterns];\n\t\t\t\t\t\tconst typeFilter = type === \"d\" ? \"d\" : type === \"l\" ? \"l\" : \"f\";\n\n\t\t\t\t\t\tconst emit = (relativized: string[]) => {\n\t\t\t\t\t\t\tif (relativized.length === 0) {\n\t\t\t\t\t\t\t\tsettle(() =>\n\t\t\t\t\t\t\t\t\tresolve({ content: [{ type: \"text\", text: NO_RESULTS_MESSAGE }], details: undefined }),\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t// Sources over-fetch by one so exactly-limit result sets are not\n\t\t\t\t\t\t\t// misreported as truncated.\n\t\t\t\t\t\t\tconst unique = [...new Set(relativized)].sort();\n\t\t\t\t\t\t\tconst resultLimitReached = unique.length > effectiveLimit;\n\t\t\t\t\t\t\tconst truncated = unique.slice(0, effectiveLimit);\n\t\t\t\t\t\t\tconst rawOutput = compress ? compressPaths(truncated) : truncated.join(\"\\n\");\n\t\t\t\t\t\t\tconst truncation = truncateHead(rawOutput, { maxLines: Number.MAX_SAFE_INTEGER });\n\t\t\t\t\t\t\tlet resultOutput = truncation.content;\n\t\t\t\t\t\t\tconst details: FindToolDetails = {};\n\t\t\t\t\t\t\tconst notices: string[] = [];\n\t\t\t\t\t\t\tif (resultLimitReached) {\n\t\t\t\t\t\t\t\tnotices.push(\n\t\t\t\t\t\t\t\t\t`${effectiveLimit} result${effectiveLimit === 1 ? \"\" : \"s\"} limit reached. Use limit=${effectiveLimit * 2} for more, or refine pattern`,\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\tdetails.resultLimitReached = effectiveLimit;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (truncation.truncated) {\n\t\t\t\t\t\t\t\tnotices.push(`${formatSize(DEFAULT_MAX_BYTES)} limit reached`);\n\t\t\t\t\t\t\t\tdetails.truncation = truncation;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (notices.length > 0) resultOutput += `\\n\\n[${notices.join(\". \")}]`;\n\t\t\t\t\t\t\tsettle(() =>\n\t\t\t\t\t\t\t\tresolve({\n\t\t\t\t\t\t\t\t\tcontent: [{ type: \"text\", text: resultOutput }],\n\t\t\t\t\t\t\t\t\tdetails: Object.keys(details).length > 0 ? details : undefined,\n\t\t\t\t\t\t\t\t}),\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t};\n\n\t\t\t\t\t\t// If custom operations provide glob(), use that instead of fd.\n\t\t\t\t\t\tif (customOps?.glob) {\n\t\t\t\t\t\t\tif (!(await ops.exists(searchPath))) {\n\t\t\t\t\t\t\t\tsettle(() => reject(new Error(`Path not found: ${searchPath}`)));\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tconst results = await ops.glob(patterns, searchPath, {\n\t\t\t\t\t\t\t\tignore: allIgnore,\n\t\t\t\t\t\t\t\tlimit: effectiveLimit + 1,\n\t\t\t\t\t\t\t\ttype: typeFilter,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\temit(\n\t\t\t\t\t\t\t\tresults.map((p) => {\n\t\t\t\t\t\t\t\t\tif (p.startsWith(searchPath)) return toPosixPath(p.slice(searchPath.length + 1));\n\t\t\t\t\t\t\t\t\treturn toPosixPath(path.relative(searchPath, p));\n\t\t\t\t\t\t\t\t}),\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\t// Default implementation uses fd, with a pure-JS fallback when fd is\n\t\t\t\t\t\t// unavailable (restricted environments) or explicitly forced.\n\t\t\t\t\t\tconst fdPath = isNativeSearchForced() ? undefined : await ensureTool(\"fd\", true);\n\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\t\t\t\t\t\tif (!fdPath) {\n\t\t\t\t\t\t\tif (!(await ops.exists(searchPath))) {\n\t\t\t\t\t\t\t\tsettle(() => reject(new Error(`Path not found: ${searchPath}`)));\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tconst nativeResults = nativeFind(searchPath, {\n\t\t\t\t\t\t\t\tpatterns,\n\t\t\t\t\t\t\t\ttype: typeFilter,\n\t\t\t\t\t\t\t\texcludeGlobs: allIgnore,\n\t\t\t\t\t\t\t\tmaxDepth: depth,\n\t\t\t\t\t\t\t\t// find always excludes node_modules/.git; the exclude globs above\n\t\t\t\t\t\t\t\t// cover them too, but skipping the dirs avoids descending them.\n\t\t\t\t\t\t\t\talwaysSkipDirs: new Set([\".git\", \"node_modules\"]),\n\t\t\t\t\t\t\t\tsignal,\n\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\temit(nativeResults);\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tconst children: ChildProcess[] = [];\n\t\t\t\t\t\tstopChildren = () => {\n\t\t\t\t\t\t\tfor (const child of children) if (!child.killed) child.kill();\n\t\t\t\t\t\t};\n\n\t\t\t\t\t\t// Run each pattern with fd. fd errors (e.g. an invalid glob) surface as a\n\t\t\t\t\t\t// rejection when the pattern produced no output, matching the shell's\n\t\t\t\t\t\t// behavior; a non-zero exit that still produced matches is tolerated.\n\t\t\t\t\t\tconst runPattern = (pattern: string): Promise<string[]> =>\n\t\t\t\t\t\t\tnew Promise<string[]>((res, rej) => {\n\t\t\t\t\t\t\t\tconst args: string[] = [\n\t\t\t\t\t\t\t\t\t\"--glob\",\n\t\t\t\t\t\t\t\t\t\"--color=never\",\n\t\t\t\t\t\t\t\t\t\"--hidden\",\n\t\t\t\t\t\t\t\t\t\"--no-require-git\",\n\t\t\t\t\t\t\t\t\t\"--type\",\n\t\t\t\t\t\t\t\t\ttypeFilter,\n\t\t\t\t\t\t\t\t\t\"--max-results\",\n\t\t\t\t\t\t\t\t\tString(effectiveLimit + 1),\n\t\t\t\t\t\t\t\t];\n\t\t\t\t\t\t\t\tfor (const excl of allIgnore) args.push(\"--exclude\", excl);\n\t\t\t\t\t\t\t\tif (depth !== undefined) args.push(\"--max-depth\", String(depth));\n\t\t\t\t\t\t\t\tconst effectivePattern = applyFdGlobPattern(args, pattern);\n\t\t\t\t\t\t\t\targs.push(\"--\", effectivePattern, searchPath);\n\n\t\t\t\t\t\t\t\tconst child = spawn(fdPath, args, { stdio: [\"ignore\", \"pipe\", \"pipe\"] });\n\t\t\t\t\t\t\t\tchildren.push(child);\n\t\t\t\t\t\t\t\tconst rl = createInterface({ input: child.stdout! });\n\t\t\t\t\t\t\t\tlet stderr = \"\";\n\t\t\t\t\t\t\t\tconst lines: string[] = [];\n\t\t\t\t\t\t\t\tchild.stderr?.on(\"data\", (chunk) => {\n\t\t\t\t\t\t\t\t\tstderr += chunk.toString();\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\trl.on(\"line\", (line) => lines.push(line));\n\t\t\t\t\t\t\t\tchild.on(\"error\", (error) => {\n\t\t\t\t\t\t\t\t\trl.close();\n\t\t\t\t\t\t\t\t\trej(new Error(`Failed to run fd: ${error.message}`));\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t\tchild.on(\"close\", (code) => {\n\t\t\t\t\t\t\t\t\trl.close();\n\t\t\t\t\t\t\t\t\tif (code !== 0 && lines.length === 0) {\n\t\t\t\t\t\t\t\t\t\trej(new Error(stderr.trim() || `fd exited with code ${code}`));\n\t\t\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t\tres(lines);\n\t\t\t\t\t\t\t\t});\n\t\t\t\t\t\t\t});\n\n\t\t\t\t\t\tconst patternResults = await Promise.all(patterns.map(runPattern));\n\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\n\t\t\t\t\t\tconst relativized: string[] = [];\n\t\t\t\t\t\tfor (const results of patternResults) {\n\t\t\t\t\t\t\tfor (const rawLine of results) {\n\t\t\t\t\t\t\t\tconst line = rawLine.replace(/\\r$/, \"\").trim();\n\t\t\t\t\t\t\t\tif (!line) continue;\n\t\t\t\t\t\t\t\trelativized.push(relativizeFdLine(line, searchPath));\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t}\n\t\t\t\t\t\temit(relativized);\n\t\t\t\t\t} catch (e) {\n\t\t\t\t\t\tif (signal?.aborted) {\n\t\t\t\t\t\t\tsettle(() => reject(new Error(\"Operation aborted\")));\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\t\t\t\t\t\tconst error = e instanceof Error ? e : new Error(String(e));\n\t\t\t\t\t\tsettle(() => reject(error));\n\t\t\t\t\t}\n\t\t\t\t})();\n\t\t\t});\n\t\t},\n\t\trenderCall(args, theme, context) {\n\t\t\tconst text = (context.lastComponent as Text | undefined) ?? new Text(\"\", 0, 0);\n\t\t\ttext.setText(formatFindCall(args, theme));\n\t\t\treturn text;\n\t\t},\n\t\trenderResult(result, options, theme, context) {\n\t\t\tconst text = (context.lastComponent as Text | undefined) ?? new Text(\"\", 0, 0);\n\t\t\ttext.setText(formatFindResult(result as any, options, theme, context.showImages));\n\t\t\treturn text;\n\t\t},\n\t};\n}\n\nexport function createFindTool(cwd: string, options?: FindToolOptions): AgentTool<typeof findSchema> {\n\treturn wrapToolDefinition(createFindToolDefinition(cwd, options));\n}\n"]}

package/dist/core/tools/find.js

@@ -181,7 +181,7 @@ export function createFindToolDefinition(cwd, options) {
                 (async () => {
                     try {
                         const searchPath = resolveToCwd(searchDir || ".", cwd);
-                        const effectiveLimit = limit ?? DEFAULT_LIMIT;
+                        const effectiveLimit = Math.max(1, limit ?? DEFAULT_LIMIT);
                         const ops = customOps ?? defaultFindOperations;
                         const patterns = Array.isArray(rawPattern) ? rawPattern : [rawPattern];
                         const excludePatterns = rawExclude ? (Array.isArray(rawExclude) ? rawExclude : [rawExclude]) : [];
@@ -193,8 +193,10 @@ export function createFindToolDefinition(cwd, options) {
                                 settle(() => resolve({ content: [{ type: "text", text: NO_RESULTS_MESSAGE }], details: undefined }));
                                 return;
                             }
+                            // Sources over-fetch by one so exactly-limit result sets are not
+                            // misreported as truncated.
                             const unique = [...new Set(relativized)].sort();
-                            const resultLimitReached = unique.length >= effectiveLimit;
+                            const resultLimitReached = unique.length > effectiveLimit;
                             const truncated = unique.slice(0, effectiveLimit);
                             const rawOutput = compress ? compressPaths(truncated) : truncated.join("\n");
                             const truncation = truncateHead(rawOutput, { maxLines: Number.MAX_SAFE_INTEGER });
@@ -202,7 +204,7 @@ export function createFindToolDefinition(cwd, options) {
                             const details = {};
                             const notices = [];
                             if (resultLimitReached) {
-                                notices.push(`${effectiveLimit} results limit reached. Use limit=${effectiveLimit * 2} for more, or refine pattern`);
+                                notices.push(`${effectiveLimit} result${effectiveLimit === 1 ? "" : "s"} limit reached. Use limit=${effectiveLimit * 2} for more, or refine pattern`);
                                 details.resultLimitReached = effectiveLimit;
                             }
                             if (truncation.truncated) {
@@ -228,7 +230,7 @@ export function createFindToolDefinition(cwd, options) {
                             }
                             const results = await ops.glob(patterns, searchPath, {
                                 ignore: allIgnore,
-                                limit: effectiveLimit,
+                                limit: effectiveLimit + 1,
                                 type: typeFilter,
                             });
                             if (signal?.aborted) {
@@ -289,7 +291,7 @@ export function createFindToolDefinition(cwd, options) {
                                 "--type",
                                 typeFilter,
                                 "--max-results",
-                                String(effectiveLimit),
+                                String(effectiveLimit + 1),
                             ];
                             for (const excl of allIgnore)
                                 args.push("--exclude", excl);