@maxgfr/codeindex 2.12.0 → 2.14.0

package/scripts/engine.mjs

@@ -14,9 +14,9 @@ var ENGINE_VERSION, SCHEMA_VERSION, EXTRACTOR_VERSION;
 var init_types = __esm({
   "src/types.ts"() {
     "use strict";
-    ENGINE_VERSION = "2.12.0";
+    ENGINE_VERSION = "2.14.0";
     SCHEMA_VERSION = 4;
-    EXTRACTOR_VERSION = 9;
+    EXTRACTOR_VERSION = 10;
   }
 });
 
@@ -320,6 +320,7 @@ function walk(root, opts = {}) {
   const maxFileBytes = opts.maxFileBytes ?? 1024 * 1024;
   const maxFiles = opts.maxFiles ?? DEFAULT_MAX_FILES;
   const useGitignore = opts.gitignore !== false;
+  const ignoreDirs = opts.ignoreDirs ? new Set(opts.ignoreDirs) : IGNORE_DIRS;
   const out2 = [];
   let capped = false;
   let excluded = 0;
@@ -347,28 +348,31 @@ function walk(root, opts = {}) {
     if (!contained(real)) continue;
     let entries;
     try {
-      entries = readdirSync(frame.dir).sort();
+      entries = readdirSync(frame.dir, { withFileTypes: true }).sort(
+        (a, b) => a.name < b.name ? -1 : a.name > b.name ? 1 : 0
+      );
     } catch {
       continue;
     }
     let rules = frame.rules;
-    if (useGitignore && entries.includes(".gitignore")) {
+    if (useGitignore && entries.some((e) => e.name === ".gitignore")) {
       const parsed = parseGitignore(readText(join(frame.dir, ".gitignore")), frame.rel);
       if (parsed.length) rules = [...rules, ...parsed];
     }
-    for (const name2 of entries) {
+    for (const entry of entries) {
+      const name2 = entry.name;
       const abs = join(frame.dir, name2);
       const rel = frame.rel ? `${frame.rel}/${name2}` : name2;
+      const isLink = entry.isSymbolicLink();
+      if (entry.isDirectory() && ignoreDirs.has(name2)) continue;
       let st;
-      let isLink;
       try {
-        st = statSync(abs);
-        isLink = lstatSync(abs).isSymbolicLink();
+        st = isLink ? statSync(abs) : lstatSync(abs);
       } catch {
         continue;
       }
       if (st.isDirectory()) {
-        if (IGNORE_DIRS.has(name2)) continue;
+        if (ignoreDirs.has(name2)) continue;
         if (isLink) continue;
         if (useGitignore && rules.length && isIgnored(rules, rel, true)) continue;
         stack.push({ dir: abs, rel, rules });
@@ -463,6 +467,7 @@ var init_walk = __esm({
       ".cache",
       "tmp",
       ".ultraindex",
+      ".codeindex",
       "Pods",
       "DerivedData",
       ".terraform",
@@ -901,7 +906,7 @@ var init_js_ts = __esm({
     RULES = [
       { re: /^\s*export\s+(?:async\s+)?function\s+(?<name>[\w$]+)/, kind: "function", exported: true },
       { re: /^\s*export\s+default\s+(?:async\s+)?function\s+(?<name>[\w$]+)/, kind: "function", exported: true },
-      { re: /^\s*export\s+default\s+(?:abstract\s+)?class\s+(?<name>[\w$]+)/, kind: "class", exported: true },
+      { re: /^\s*export\s+default\s+(?:abstract\s+)?class\s+(?!extends\b)(?<name>[\w$]+)/, kind: "class", exported: true },
       { re: /^\s*(?:async\s+)?function\s+(?<name>[\w$]+)/, kind: "function", exported: false },
       { re: /^\s*export\s+(?:abstract\s+)?class\s+(?<name>[\w$]+)/, kind: "class", exported: true },
       { re: /^\s*(?:abstract\s+)?class\s+(?<name>[\w$]+)/, kind: "class", exported: false },
@@ -1770,13 +1775,13 @@ async function Module2(moduleArg = {}) {
       }
       readAsync = /* @__PURE__ */ __name(async (url) => {
         if (isFileURI(url)) {
-          return new Promise((resolve2, reject) => {
+          return new Promise((resolve3, reject) => {
             var xhr = new XMLHttpRequest();
             xhr.open("GET", url, true);
             xhr.responseType = "arraybuffer";
             xhr.onload = () => {
               if (xhr.status == 200 || xhr.status == 0 && xhr.response) {
-                resolve2(xhr.response);
+                resolve3(xhr.response);
                 return;
               }
               reject(xhr.status);
@@ -1972,9 +1977,9 @@ async function Module2(moduleArg = {}) {
     __name(receiveInstantiationResult, "receiveInstantiationResult");
     var info2 = getWasmImports();
     if (Module["instantiateWasm"]) {
-      return new Promise((resolve2, reject) => {
+      return new Promise((resolve3, reject) => {
         Module["instantiateWasm"](info2, (mod, inst) => {
-          resolve2(receiveInstance(mod, inst));
+          resolve3(receiveInstance(mod, inst));
         });
       });
     }
@@ -3305,8 +3310,8 @@ async function Module2(moduleArg = {}) {
   if (runtimeInitialized) {
     moduleRtn = Module;
   } else {
-    moduleRtn = new Promise((resolve2, reject) => {
-      readyPromiseResolve = resolve2;
+    moduleRtn = new Promise((resolve3, reject) => {
+      readyPromiseResolve = resolve3;
       readyPromiseReject = reject;
     });
   }
@@ -5537,27 +5542,41 @@ ${JSON.stringify(symbolNames, null, 2)}`);
 
 // src/ast/loader.ts
 import { readFileSync as readFileSync2, existsSync } from "fs";
+import { homedir } from "os";
 import { dirname, join as join2 } from "path";
 import { fileURLToPath } from "url";
 function grammarKeyForExt(ext) {
   return EXT_GRAMMAR[ext];
 }
-function resolveGrammarDir() {
-  const env = process.env.CODEINDEX_GRAMMAR_DIR ?? process.env.ULTRAINDEX_GRAMMAR_DIR;
-  if (env && existsSync(env)) return env;
-  const here = dirname(fileURLToPath(import.meta.url));
-  const candidates = [
+function sharedGrammarsCacheDir() {
+  const xdg = process.env.XDG_CACHE_HOME;
+  const base = xdg && xdg.trim() ? xdg.trim() : join2(homedir(), ".cache");
+  return join2(base, "codeindex", "grammars", ENGINE_VERSION);
+}
+function resolveGrammarsTier(opts = {}) {
+  const cacheDir = sharedGrammarsCacheDir();
+  const legacy = process.env.CODEINDEX_GRAMMAR_DIR ?? process.env.ULTRAINDEX_GRAMMAR_DIR;
+  if (legacy && legacy.trim() && existsSync(legacy)) return { tier: "env", dir: legacy, cacheDir };
+  const here = opts.moduleDir ?? dirname(fileURLToPath(import.meta.url));
+  const adjacent = [
     join2(here, "grammars"),
     // bundle: <...>/scripts/grammars
     join2(here, "..", "..", "scripts", "grammars"),
     // dev: src/ast → <repo>/scripts/grammars
     join2(here, "..", "scripts", "grammars")
   ];
-  for (const c2 of candidates) if (existsSync(c2)) return c2;
-  return join2(here, "grammars");
+  for (const c2 of adjacent) if (existsSync(c2)) return { tier: "adjacent", dir: c2, cacheDir };
+  const env = process.env.CODEINDEX_GRAMMARS_DIR;
+  if (env && env.trim() && existsSync(env)) return { tier: "env", dir: env, cacheDir };
+  if (existsSync(cacheDir)) return { tier: "cache", dir: cacheDir, cacheDir };
+  return { tier: "none", cacheDir };
+}
+function resolveGrammarsDir(opts) {
+  return resolveGrammarsTier(opts).dir;
 }
 async function ensureGrammars(keys) {
-  const dir = resolveGrammarDir();
+  const dir = resolveGrammarsDir();
+  if (!dir) return;
   if (!runtimeReady) {
     const runtime = join2(dir, "web-tree-sitter.wasm");
     if (!existsSync(runtime)) return;
@@ -5582,6 +5601,14 @@ async function ensureGrammars(keys) {
 function allGrammarKeys() {
   return [...new Set(Object.values(EXT_GRAMMAR))];
 }
+function grammarKeysForExts(exts) {
+  const keys = /* @__PURE__ */ new Set();
+  for (const ext of exts) {
+    const key = EXT_GRAMMAR[ext];
+    if (key !== void 0) keys.add(key);
+  }
+  return [...keys].sort();
+}
 function grammarReady(key) {
   return loaded.has(key);
 }
@@ -5596,6 +5623,7 @@ var init_loader = __esm({
   "src/ast/loader.ts"() {
     "use strict";
     init_web_tree_sitter();
+    init_types();
     EXT_GRAMMAR = {
       ".ts": "typescript",
       ".mts": "typescript",
@@ -5718,7 +5746,7 @@ function readReceiver(node) {
   const name2 = obj ? readName(obj) : void 0;
   return name2 && /^[A-Za-z_]\w*$/.test(name2) ? name2 : void 0;
 }
-function collectCalls(root, spec) {
+function collectCalls(root, spec, maxCalls = MAX_CALLS) {
   if (!spec.calls) return [];
   const out2 = [];
   const seen = /* @__PURE__ */ new Set();
@@ -5749,7 +5777,7 @@ function collectCalls(root, spec) {
   };
   visit(root);
   out2.sort((a, b) => byStr(a.name, b.name) || a.line - b.line);
-  return out2.slice(0, MAX_CALLS);
+  return out2.slice(0, maxCalls);
 }
 function collectImportedNames(root, spec) {
   if (!spec.imports?.import_statement) return [];
@@ -5776,7 +5804,7 @@ function collectImportedNames(root, spec) {
   visit(root);
   return [...found].sort(byStr).slice(0, MAX_IMPORTED_NAMES);
 }
-function extractAst(rel, ext, content) {
+function extractAst(rel, ext, content, opts = {}) {
   const key = grammarKeyForExt(ext);
   if (!key || !grammarReady(key)) return void 0;
   const spec = SPECS[key];
@@ -5962,7 +5990,7 @@ function extractAst(rel, ext, content) {
     }
     const refs = collectImports(root, spec);
     const idents = collectRefIdents(root, new Set(symbols.map((s) => s.name)));
-    const calls = collectCalls(root, spec);
+    const calls = collectCalls(root, spec, opts.maxCalls);
     const importedNames = collectImportedNames(root, spec);
     let pkg;
     if (spec.lang === "java") {
@@ -6398,12 +6426,12 @@ function extractImports(ext, content) {
   }
   return [...specs].map((spec) => ({ kind: "import", spec }));
 }
-function collectCallsRegex(content, symbols = []) {
+function collectCallsRegex(content, symbols = [], maxCalls = 512) {
   const out2 = /* @__PURE__ */ new Map();
   const ownDefLines = new Set(symbols.map((s) => `${s.name} ${s.line}`));
   const lines = content.split("\n");
   const CALL_RE = /(?:\bnew\s+)?(?:([A-Za-z_$][\w$]*)\s*\.\s*)?([A-Za-z_$][\w$]*)\s*\(/g;
-  for (let i2 = 0; i2 < lines.length && out2.size < 512; i2++) {
+  for (let i2 = 0; i2 < lines.length && out2.size < maxCalls; i2++) {
     const line = lines[i2];
     const trimmed = line.trimStart();
     if (trimmed.startsWith("//") || trimmed.startsWith("#") || trimmed.startsWith("*")) continue;
@@ -6418,7 +6446,7 @@ function collectCallsRegex(content, symbols = []) {
     CALL_RE.lastIndex = 0;
     let m;
     const fallbackExcluded = /* @__PURE__ */ new Set();
-    while ((m = CALL_RE.exec(line)) !== null && out2.size < 512) {
+    while ((m = CALL_RE.exec(line)) !== null && out2.size < maxCalls) {
       const receiver = m[1];
       const name2 = m[2];
       if (name2.length < 2 || CALL_KEYWORDS.has(name2)) continue;
@@ -6435,8 +6463,8 @@ function collectCallsRegex(content, symbols = []) {
   }
   return [...out2.values()].sort((a, b) => a.name < b.name ? -1 : a.name > b.name ? 1 : a.line - b.line);
 }
-function extractCode(rel, ext, content) {
-  const ast = extractAst(rel, ext, content);
+function extractCode(rel, ext, content, opts = {}) {
+  const ast = extractAst(rel, ext, content, { maxCalls: opts.maxCallsPerFile });
   const symbols = (ast ? ast.symbols : extractSymbols(rel, ext, content)).slice(0, 400);
   const known = new Set(symbols.map((s) => s.name));
   const reexports = extractReexports(rel, content, symbols).filter((s) => !known.has(s.name));
@@ -6452,7 +6480,7 @@ function extractCode(rel, ext, content) {
     // collector otherwise, so caller indexes exist without the wasm sidecar.
     // `symbols` (this file's own regex-extracted defs) lets the collector
     // exclude a definition's own name+line from its call candidates.
-    calls: ast ? ast.calls : collectCallsRegex(content, symbols),
+    calls: ast ? ast.calls : collectCallsRegex(content, symbols, opts.maxCallsPerFile),
     importedNames: ast?.importedNames
   };
 }
@@ -6521,16 +6549,20 @@ function scanRepo(root, opts = {}) {
   const scoped = opts.scope ? [...opts.include ?? [], `${opts.scope.replace(/\/+$/, "")}/**`] : opts.include;
   const include = compileGlobs(scoped);
   const exclude = compileGlobs(opts.exclude);
-  const { files: walked, capped, excluded } = walk(root, {
+  const { files: walked, capped, excluded } = opts.precomputedWalk ?? walk(root, {
     maxFileBytes: opts.maxBytes,
     maxFiles: opts.maxFiles,
-    gitignore: opts.gitignore
+    gitignore: opts.gitignore,
+    ignoreDirs: opts.ignoreDirs
   });
   const outPrefix = opts.out ? opts.out.replace(/\/+$/, "") + "/" : null;
   const files = [];
   const languages = {};
   const docText = /* @__PURE__ */ new Map();
   const mtimes = /* @__PURE__ */ new Map();
+  const cache = opts.cache;
+  let allReused = cache !== void 0;
+  let cacheDirty = cache === void 0;
   for (const f of walked) {
     if (outPrefix && (f.abs === opts.out || f.abs.startsWith(outPrefix))) continue;
     if (include && !include(f.rel)) continue;
@@ -6549,8 +6581,11 @@ function scanRepo(root, opts = {}) {
     if (cached && cached.hash === hash) {
       files.push(cached.record);
       if (kind === "doc" && content) docText.set(f.rel, content);
+      if (cached.size !== f.size || cached.mtimeMs !== f.mtimeMs) cacheDirty = true;
       continue;
     }
+    allReused = false;
+    cacheDirty = true;
     const record = {
       rel: f.rel,
       ext: f.ext,
@@ -6573,7 +6608,7 @@ function scanRepo(root, opts = {}) {
       } else if (kind === "doc") {
         record.title = basename(f.rel);
       } else if (kind === "code") {
-        const code = extractCode(f.rel, f.ext, content);
+        const code = extractCode(f.rel, f.ext, content, { maxCallsPerFile: opts.maxCallsPerFile });
         record.title = basename(f.rel);
         record.summary = code.summary;
         record.symbols = code.symbols;
@@ -6592,7 +6627,22 @@ function scanRepo(root, opts = {}) {
     files.push(record);
   }
   files.sort(byKey((f) => f.rel));
-  return { root, commit: headCommit(root), files, languages, docText, mtimes, capped, excluded };
+  if (cache !== void 0 && files.length !== cache.size) {
+    allReused = false;
+    cacheDirty = true;
+  }
+  return {
+    root,
+    commit: headCommit(root),
+    files,
+    languages,
+    docText,
+    mtimes,
+    capped,
+    excluded,
+    contentUnchanged: allReused,
+    cacheDirty
+  };
 }
 var init_scan = __esm({
   "src/scan.ts"() {
@@ -6611,7 +6661,7 @@ var init_scan = __esm({
 
 // src/resolve.ts
 import { posix } from "path";
-import { join as join3 } from "path";
+import { join as join4 } from "path";
 function distToSrcCandidates(target) {
   const segs = norm(target).split("/").filter((s) => s !== ".");
   const out2 = [];
@@ -6698,7 +6748,7 @@ function resolveExtends(fileSet, fromDir, ext) {
 function readTsConfig(root, fileSet, rel, warnings, seen) {
   if (seen.has(rel)) return void 0;
   seen.add(rel);
-  const cfg = tolerantJsonParse(readText(join3(root, rel)));
+  const cfg = tolerantJsonParse(readText(join4(root, rel)));
   if (cfg === void 0) {
     warnings.push(`unparseable ${rel} \u2014 its path aliases were ignored`);
     return void 0;
@@ -6827,7 +6877,7 @@ function buildResolveContext(scan2) {
   const goModules = [];
   for (const rel of fileSet) {
     if (rel !== "go.mod" && !rel.endsWith("/go.mod")) continue;
-    const text = readText(join3(scan2.root, rel));
+    const text = readText(join4(scan2.root, rel));
     const m = /^\s*module\s+(\S+)/m.exec(text);
     if (!m) continue;
     const dir = rel.includes("/") ? posix.dirname(rel) : "";
@@ -6837,7 +6887,7 @@ function buildResolveContext(scan2) {
   const rustCrates = [];
   for (const rel of fileSet) {
     if (rel !== "Cargo.toml" && !rel.endsWith("/Cargo.toml")) continue;
-    const text = readText(join3(scan2.root, rel));
+    const text = readText(join4(scan2.root, rel));
     const m = /\[package\][^[]*?^\s*name\s*=\s*"([^"]+)"/ms.exec(text);
     if (!m) continue;
     const dir = rel.includes("/") ? posix.dirname(rel) : "";
@@ -6864,7 +6914,7 @@ function buildResolveContext(scan2) {
   const workspacePackages = [];
   for (const rel of fileSet) {
     if (rel !== "package.json" && !rel.endsWith("/package.json")) continue;
-    const pkg = tolerantJsonParse(readText(join3(scan2.root, rel)));
+    const pkg = tolerantJsonParse(readText(join4(scan2.root, rel)));
     if (pkg === void 0) {
       warnings.push(`unparseable ${rel} \u2014 skipped for workspace resolution`);
       continue;
@@ -6891,7 +6941,7 @@ function buildResolveContext(scan2) {
   const phpPsr4 = [];
   for (const rel of fileSet) {
     if (rel !== "composer.json" && !rel.endsWith("/composer.json")) continue;
-    const composer = tolerantJsonParse(readText(join3(scan2.root, rel)));
+    const composer = tolerantJsonParse(readText(join4(scan2.root, rel)));
     if (!composer) {
       warnings.push(`unparseable ${rel} \u2014 skipped for PHP PSR-4 resolution`);
       continue;
@@ -7466,7 +7516,7 @@ var init_calls = __esm({
 });
 
 // src/graph.ts
-import { join as join4 } from "path";
+import { join as join5 } from "path";
 function isDistinctive(name2) {
   if (name2.length < 5) return false;
   const internalUpper = /[a-z][A-Z]/.test(name2) || /[A-Z]{2}/.test(name2);
@@ -7545,7 +7595,7 @@ function buildGraph(scan2, ctx, modules, moduleOf, meta) {
   if (unique.size) {
     for (const f of scan2.files) {
       if (f.kind !== "doc") continue;
-      const content = scan2.docText.get(f.rel) ?? readText(join4(scan2.root, f.rel));
+      const content = scan2.docText.get(f.rel) ?? readText(join5(scan2.root, f.rel));
       if (!content) continue;
       const tokens = /* @__PURE__ */ new Map();
       for (const tok of content.split(/[^A-Za-z0-9_]+/)) {
@@ -7650,217 +7700,567 @@ var init_graph = __esm({
   }
 });
 
-// src/callers.ts
-function computeImportPairs(scan2) {
-  const ctx = buildResolveContext(scan2);
-  const pairs = /* @__PURE__ */ new Set();
+// src/render/symbols-json.ts
+function computeSymbolRefs(scan2) {
+  const unique = uniqueSymbolDefs(scan2);
+  const refs = /* @__PURE__ */ new Map();
+  if (!unique.size) return refs;
+  const add = (name2, file) => {
+    let set = refs.get(name2);
+    if (!set) refs.set(name2, set = /* @__PURE__ */ new Set());
+    set.add(file);
+  };
   for (const f of scan2.files) {
-    for (const ref of f.refs) {
-      if (ref.kind !== "import") continue;
-      const r = resolveImport(f.rel, f.ext, ref.spec, ctx);
-      if (r.kind === "resolved" && r.target !== f.rel) pairs.add(`${f.rel}|${r.target}`);
+    if (f.kind === "code" && f.idents) {
+      for (const id of f.idents) {
+        const target = unique.get(id);
+        if (target && target !== f.rel) add(id, f.rel);
+      }
+    } else if (f.kind === "doc") {
+      const content = scan2.docText.get(f.rel);
+      if (!content) continue;
+      for (const tok of content.split(/[^A-Za-z0-9_]+/)) {
+        const target = unique.get(tok);
+        if (target && target !== f.rel) add(tok, f.rel);
+      }
     }
   }
-  return pairs;
+  return refs;
 }
-function buildCallerIndex(scan2, importPairs, opts = {}) {
-  const pairs = importPairs ?? computeImportPairs(scan2);
-  const recall = opts.recall === true;
-  const defs = /* @__PURE__ */ new Map();
-  for (const f of scan2.files) {
-    const seen = /* @__PURE__ */ new Set();
-    for (const s of f.symbols) {
-      if (!s.exported || REFERENCE_KINDS3.has(s.kind)) continue;
-      if (seen.has(s.name)) continue;
-      seen.add(s.name);
-      let arr = defs.get(s.name);
-      if (!arr) defs.set(s.name, arr = []);
-      arr.push(s);
-    }
-  }
-  const localDefs = /* @__PURE__ */ new Map();
+function buildSymbolIndex(scan2, refs = /* @__PURE__ */ new Map()) {
+  const defsByName = /* @__PURE__ */ new Map();
   for (const f of scan2.files) {
-    const byName = /* @__PURE__ */ new Map();
     for (const s of f.symbols) {
-      if (!REFERENCE_KINDS3.has(s.kind) && !byName.has(s.name)) byName.set(s.name, s);
-    }
-    localDefs.set(f.rel, byName);
-  }
-  const sites = /* @__PURE__ */ new Map();
-  const record = (def, caller) => {
-    let entry = sites.get(def.name + "\0" + def.file);
-    if (!entry) sites.set(def.name + "\0" + def.file, entry = { def, callers: [] });
-    entry.callers.push(caller);
-  };
-  for (const f of scan2.files) {
-    if (!f.calls?.length) continue;
-    const family = familyOf(f.lang);
-    const own = localDefs.get(f.rel);
-    for (const c2 of f.calls) {
-      const local = own.get(c2.name);
-      if (local) {
-        if (local.line !== c2.line)
-          record(local, recall ? { file: f.rel, line: c2.line, confidence: "corroborated" } : { file: f.rel, line: c2.line });
-        continue;
-      }
-      const cands = (defs.get(c2.name) ?? []).filter((d) => familyOf(d.lang) === family && d.file !== f.rel).map((d) => ({ file: d.file, lang: d.lang }));
-      if (!cands.length) continue;
-      const imported = cands.filter((d) => pairs.has(`${f.rel}|${d.file}`));
-      const chosen = family === "js" ? imported.length ? pickCandidate(f.rel, imported) : (
-        // JS/TS gate: no corroborating import → no binding. Recall mode
-        // relaxes this to a unique-repo-wide name match (issue #7).
-        recall && cands.length === 1 ? cands[0] : void 0
-      ) : imported.length ? pickCandidate(f.rel, imported) : pickCandidate(f.rel, cands);
-      if (!chosen) continue;
-      const def = defs.get(c2.name).find((d) => d.file === chosen.file);
-      record(
-        def,
-        recall ? { file: f.rel, line: c2.line, confidence: imported.length ? "corroborated" : "unique-name" } : { file: f.rel, line: c2.line }
-      );
+      let arr = defsByName.get(s.name);
+      if (!arr) defsByName.set(s.name, arr = []);
+      arr.push({
+        file: s.file,
+        line: s.line,
+        ...s.endLine !== void 0 ? { endLine: s.endLine } : {},
+        kind: s.kind,
+        exported: s.exported,
+        lang: s.lang,
+        ...s.parent ? { parent: s.parent } : {}
+      });
     }
   }
-  const index = /* @__PURE__ */ new Map();
-  const keys = [...sites.keys()].sort(byStr);
-  for (const key of keys) {
-    const { def, callers } = sites.get(key);
-    callers.sort((a, b) => byStr(a.file, b.file) || a.line - b.line);
-    if (!index.has(def.name)) index.set(def.name, { def, callers });
-    else index.set(`${def.name}@${def.file}`, { def, callers });
+  const defs = {};
+  for (const name2 of [...defsByName.keys()].sort(byStr)) {
+    defs[name2] = defsByName.get(name2).slice().sort((a, b) => byStr(a.file, b.file) || a.line - b.line || byStr(a.kind, b.kind));
   }
-  return index;
-}
-function enclosingSymbol(scan2, file, line) {
-  const f = scan2.files.find((x) => x.rel === file);
-  if (!f?.symbols.length) return void 0;
-  return enclosingAmong(f.symbols, line);
-}
-function enclosingAmong(symbols, line) {
-  let best;
-  for (const s of symbols) {
-    if (REFERENCE_KINDS3.has(s.kind)) continue;
-    if (s.line > line) continue;
-    if (s.endLine !== void 0 && line > s.endLine) continue;
-    if (!best || s.line > best.line || s.line === best.line && (s.endLine ?? Infinity) <= (best.endLine ?? Infinity)) {
-      best = s;
-    }
+  const refsOut = {};
+  for (const name2 of [...refs.keys()].sort(byStr)) {
+    const files = [...refs.get(name2)].sort(byStr);
+    if (files.length) refsOut[name2] = files;
   }
-  return best;
+  return { schemaVersion: SCHEMA_VERSION, defs, refs: refsOut };
 }
-function buildRawCallerIndex(scan2) {
-  const byName = /* @__PURE__ */ new Map();
-  for (const f of scan2.files) {
-    if (!f.calls?.length) continue;
-    const symbols = f.symbols.filter((s) => !REFERENCE_KINDS3.has(s.kind));
-    for (const c2 of f.calls) {
-      const site = { file: f.rel, line: c2.line };
-      if (c2.receiver !== void 0) site.receiver = c2.receiver;
-      const enc = enclosingAmong(symbols, c2.line);
-      if (enc) site.enclosingSymbol = enc;
-      let arr = byName.get(c2.name);
-      if (!arr) byName.set(c2.name, arr = []);
-      arr.push(site);
-    }
-  }
-  const index = /* @__PURE__ */ new Map();
-  for (const name2 of [...byName.keys()].sort(byStr)) {
-    const sites = byName.get(name2);
-    sites.sort((a, b) => byStr(a.file, b.file) || a.line - b.line);
-    index.set(name2, sites);
-  }
-  return index;
+function renderSymbolsJson(index) {
+  return JSON.stringify(index, null, 2) + "\n";
 }
-var REFERENCE_KINDS3;
-var init_callers = __esm({
-  "src/callers.ts"() {
+var init_symbols_json = __esm({
+  "src/render/symbols-json.ts"() {
     "use strict";
-    init_calls();
-    init_resolve();
+    init_types();
     init_sort();
-    REFERENCE_KINDS3 = /* @__PURE__ */ new Set(["reexport", "reexport-all", "default"]);
+    init_graph();
   }
 });
 
-// src/query.ts
-import { join as join5 } from "path";
-function symbolsOverview(scan2, rel) {
-  const f = scan2.files.find((x) => x.rel === rel);
-  if (!f) return [];
-  return [...f.symbols].filter((s) => !REFERENCE_KINDS4.has(s.kind)).sort((a, b) => a.line - b.line || byStr(a.name, b.name));
-}
-function findSymbol(scan2, namePath, opts = {}) {
-  const segments = namePath.split("/").filter(Boolean);
-  if (!segments.length) return [];
-  const leaf = segments[segments.length - 1];
-  const parents = segments.slice(0, -1);
-  const matchName = (name2, wanted) => opts.substring ? name2.toLowerCase().includes(wanted.toLowerCase()) : name2 === wanted;
+// src/bm25.ts
+function subtokens(raw) {
+  const folded = foldText(raw).replace(/([a-z0-9])([A-Z])/g, "$1 $2").replace(/([A-Z]+)([A-Z][a-z])/g, "$1 $2");
   const out2 = [];
-  for (const f of scan2.files) {
-    for (const s of f.symbols) {
-      if (REFERENCE_KINDS4.has(s.kind)) continue;
-      if (!matchName(s.name, leaf)) continue;
-      if (parents.length) {
-        const parent = parents[parents.length - 1];
-        if (!s.parent || s.parent !== parent) continue;
-      }
-      out2.push({ ...s });
-    }
-  }
-  out2.sort(
-    (a, b) => Number(b.name === leaf) - Number(a.name === leaf) || byStr(a.file, b.file) || a.line - b.line
-  );
-  const capped = out2.slice(0, opts.maxResults ?? 50);
-  if (opts.includeBody) {
-    for (const m of capped) {
-      const end = m.endLine ?? m.line;
-      const content = readText(join5(scan2.root, m.file));
-      if (!content) continue;
-      m.body = content.split("\n").slice(m.line - 1, end).join("\n");
-    }
+  const seen = /* @__PURE__ */ new Set();
+  const push = (t) => {
+    if (t.length < 2 || seen.has(t)) return;
+    seen.add(t);
+    out2.push(t);
+  };
+  if (!/\s/.test(raw.trim())) push(foldText(raw).toLowerCase().replace(/[^a-z0-9_]+/g, ""));
+  for (const part of folded.split(/[^A-Za-z0-9]+/)) push(part.toLowerCase());
+  return out2;
+}
+function addTerms(doc, text) {
+  for (const t of subtokens(text)) {
+    doc.tf.set(t, (doc.tf.get(t) ?? 0) + 1);
+    doc.len++;
   }
-  return capped;
 }
-function findReferences(scan2, name2) {
-  const defs = [];
+function buildDocs(scan2) {
+  const docs = [];
   for (const f of scan2.files) {
+    const doc = { file: f.rel, tf: /* @__PURE__ */ new Map(), len: 0, symbols: [] };
+    const seenSym = /* @__PURE__ */ new Set();
     for (const s of f.symbols) {
-      if (s.name === name2 && !REFERENCE_KINDS4.has(s.kind)) defs.push(s);
-    }
-  }
-  defs.sort((a, b) => byStr(a.file, b.file) || a.line - b.line);
-  const index = buildCallerIndex(scan2);
-  const entry = index.get(name2);
-  const callSites = entry ? entry.callers : [];
-  const referencingFiles = /* @__PURE__ */ new Set();
-  const unique = uniqueSymbolDefs(scan2);
-  const defFile = unique.get(name2);
-  for (const f of scan2.files) {
-    if (f.rel === defFile) continue;
-    if (f.kind === "code" && f.idents?.includes(name2)) referencingFiles.add(f.rel);
-    else if (f.kind === "doc") {
-      const content = scan2.docText.get(f.rel);
-      if (content && new RegExp(`\\b${name2.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}\\b`).test(content)) {
-        referencingFiles.add(f.rel);
+      addTerms(doc, s.name);
+      if (!seenSym.has(s.name)) {
+        seenSym.add(s.name);
+        doc.symbols.push(s.name);
       }
     }
+    for (const seg of f.rel.split("/")) addTerms(doc, seg);
+    for (const h of f.headings) addTerms(doc, h);
+    if (f.summary) addTerms(doc, f.summary);
+    docs.push(doc);
   }
-  for (const site of callSites) referencingFiles.add(site.file);
-  return { defs, callSites, referencingFiles: [...referencingFiles].sort(byStr) };
+  return docs;
 }
-var REFERENCE_KINDS4;
-var init_query = __esm({
-  "src/query.ts"() {
+function charTrigrams(term) {
+  const padded = `^^${term}$$`;
+  const grams = /* @__PURE__ */ new Set();
+  for (let i2 = 0; i2 + 3 <= padded.length; i2++) grams.add(padded.slice(i2, i2 + 3));
+  return grams;
+}
+function diceCoefficient(a, b) {
+  if (!a.size || !b.size) return 0;
+  let inter = 0;
+  for (const g of a) if (b.has(g)) inter++;
+  return 2 * inter / (a.size + b.size);
+}
+function buildTrigramIndex(docs) {
+  const index = /* @__PURE__ */ new Map();
+  for (const d of docs) {
+    for (const term of d.tf.keys()) {
+      if (!index.has(term)) index.set(term, charTrigrams(term));
+    }
+  }
+  return index;
+}
+function searchIndex(scan2, query, opts = {}) {
+  const terms = [];
+  const seen = /* @__PURE__ */ new Set();
+  for (const kw of keywords(query)) {
+    for (const t of subtokens(kw)) {
+      if (seen.has(t)) continue;
+      seen.add(t);
+      terms.push(t);
+    }
+  }
+  if (!terms.length) return [];
+  const docs = bm25DocsFor(scan2);
+  const n = docs.length;
+  if (!n) return [];
+  let totalLen = 0;
+  for (const d of docs) totalLen += d.len;
+  const avgLen = totalLen / n || 1;
+  const df = /* @__PURE__ */ new Map();
+  for (const t of terms) {
+    let count = 0;
+    for (const d of docs) if (d.tf.has(t)) count++;
+    df.set(t, count);
+  }
+  const fuzzyEnabled = opts.fuzzy ?? true;
+  const fuzzyCandidates = /* @__PURE__ */ new Map();
+  if (fuzzyEnabled) {
+    const unmatched = terms.filter((t) => df.get(t) === 0);
+    if (unmatched.length) {
+      const trigramIndex = bm25TrigramsFor(scan2);
+      for (const t of unmatched) {
+        const grams = charTrigrams(t);
+        const candidates = [];
+        for (const [vocabTerm, vocabGrams] of trigramIndex) {
+          const dice = diceCoefficient(grams, vocabGrams);
+          if (dice >= FUZZY_DICE_THRESHOLD) candidates.push({ term: vocabTerm, dice });
+        }
+        candidates.sort((a, b) => b.dice - a.dice || byStr(a.term, b.term));
+        fuzzyCandidates.set(t, candidates.slice(0, FUZZY_CAP));
+      }
+    }
+  }
+  const vocabDf = /* @__PURE__ */ new Map();
+  const dfOfVocabTerm = (term) => {
+    const known = df.get(term) ?? vocabDf.get(term);
+    if (known !== void 0) return known;
+    let count = 0;
+    for (const d of docs) if (d.tf.has(term)) count++;
+    vocabDf.set(term, count);
+    return count;
+  };
+  const results = [];
+  for (const d of docs) {
+    let score = 0;
+    const matched = [];
+    const symbolTerms = /* @__PURE__ */ new Set();
+    const fuzzyHit = /* @__PURE__ */ new Set();
+    for (const t of terms) {
+      const tf = d.tf.get(t);
+      if (tf) {
+        matched.push(t);
+        symbolTerms.add(t);
+        const idf = Math.log(1 + (n - df.get(t) + 0.5) / (df.get(t) + 0.5));
+        score += idf * (tf * (K1 + 1)) / (tf + K1 * (1 - B + B * d.len / avgLen));
+        continue;
+      }
+      const candidates = fuzzyCandidates.get(t);
+      if (!candidates) continue;
+      for (const cand of candidates) {
+        const ctf = d.tf.get(cand.term);
+        if (!ctf) continue;
+        const cdf = dfOfVocabTerm(cand.term);
+        const idf = Math.log(1 + (n - cdf + 0.5) / (cdf + 0.5));
+        const contribution = idf * (ctf * (K1 + 1)) / (ctf + K1 * (1 - B + B * d.len / avgLen));
+        score += contribution * cand.dice;
+        symbolTerms.add(cand.term);
+        fuzzyHit.add(t);
+      }
+    }
+    if (!matched.length && !fuzzyHit.size) continue;
+    const scored = d.symbols.map((name2) => {
+      const toks = new Set(subtokens(name2));
+      let hits = 0;
+      for (const t of symbolTerms) if (toks.has(t)) hits++;
+      return { name: name2, hits };
+    }).filter((s) => s.hits > 0).sort((a, b) => b.hits - a.hits || byStr(a.name, b.name));
+    const result = {
+      file: d.file,
+      score: Number(score.toFixed(4)),
+      matchedTerms: matched.sort(byStr),
+      topSymbols: scored.slice(0, TOP_SYMBOLS).map((s) => s.name)
+    };
+    if (fuzzyHit.size) result.fuzzyTerms = [...fuzzyHit].sort(byStr);
+    results.push(result);
+  }
+  results.sort((a, b) => b.score - a.score || byStr(a.file, b.file));
+  return results.slice(0, opts.limit ?? DEFAULT_LIMIT);
+}
+var K1, B, DEFAULT_LIMIT, TOP_SYMBOLS, FUZZY_DICE_THRESHOLD, FUZZY_CAP;
+var init_bm25 = __esm({
+  "src/bm25.ts"() {
     "use strict";
+    init_derived();
+    init_util();
+    init_sort();
+    K1 = 1.2;
+    B = 0.75;
+    DEFAULT_LIMIT = 20;
+    TOP_SYMBOLS = 5;
+    FUZZY_DICE_THRESHOLD = 0.6;
+    FUZZY_CAP = 3;
+  }
+});
+
+// src/complexity.ts
+import { join as join6 } from "path";
+function complexityOfSource(source) {
+  return 1 + (source.match(BRANCH_RE) ?? []).length;
+}
+function symbolComplexity(scan2, rel, top = 50) {
+  const out2 = [];
+  for (const f of scan2.files) {
+    if (f.kind !== "code") continue;
+    if (rel && f.rel !== rel) continue;
+    if (!f.symbols.length) continue;
+    const lines = readText(join6(scan2.root, f.rel)).split("\n");
+    for (const s of f.symbols) {
+      if (s.kind === "reexport" || s.kind === "reexport-all") continue;
+      const end = s.endLine ?? s.line;
+      const body2 = lines.slice(s.line - 1, end).join("\n");
+      const entry = { file: f.rel, name: s.name, line: s.line, complexity: complexityOfSource(body2) };
+      if (s.endLine !== void 0) entry.endLine = s.endLine;
+      out2.push(entry);
+    }
+  }
+  out2.sort((a, b) => b.complexity - a.complexity || byStr(a.file, b.file) || a.line - b.line);
+  return out2.slice(0, top);
+}
+function riskHotspots(scan2, churn, top = 20) {
+  const complexityByFile = fileComplexityFor(scan2);
+  const out2 = scan2.files.filter((f) => f.kind === "code").map((f) => {
+    const complexity = complexityByFile.get(f.rel);
+    const commits = churn.get(f.rel) ?? 0;
+    return { file: f.rel, complexity, commits, score: (commits + 1) * complexity };
+  });
+  out2.sort((a, b) => b.score - a.score || byStr(a.file, b.file));
+  return out2.slice(0, top);
+}
+var BRANCH_RE;
+var init_complexity = __esm({
+  "src/complexity.ts"() {
+    "use strict";
+    init_derived();
     init_walk();
-    init_callers();
+    init_sort();
+    BRANCH_RE = /\b(if|elif|elsif|else\s+if|for|foreach|while|until|unless|case|when|match|catch|rescue|except)\b|&&|\|\||(?<![?:])\?(?![?.:])/g;
+  }
+});
+
+// src/derived.ts
+import { join as join7 } from "path";
+function cacheFor(scan2) {
+  let c2 = caches.get(scan2);
+  if (!c2) caches.set(scan2, c2 = {});
+  return c2;
+}
+function resolveContextFor(scan2) {
+  const c2 = cacheFor(scan2);
+  return c2.resolveCtx ??= buildResolveContext(scan2);
+}
+function importPairsFor(scan2) {
+  const c2 = cacheFor(scan2);
+  if (!c2.importPairs) {
+    const ctx = resolveContextFor(scan2);
+    const pairs = /* @__PURE__ */ new Set();
+    for (const f of scan2.files) {
+      for (const ref of f.refs) {
+        if (ref.kind !== "import") continue;
+        const r = resolveImport(f.rel, f.ext, ref.spec, ctx);
+        if (r.kind === "resolved" && r.target !== f.rel) pairs.add(`${f.rel}|${r.target}`);
+      }
+    }
+    c2.importPairs = pairs;
+  }
+  return c2.importPairs;
+}
+function uniqueDefsFor(scan2) {
+  const c2 = cacheFor(scan2);
+  return c2.uniqueDefs ??= uniqueSymbolDefs(scan2);
+}
+function symbolRefsFor(scan2) {
+  const c2 = cacheFor(scan2);
+  return c2.symbolRefs ??= computeSymbolRefs(scan2);
+}
+function callerIndexFor(scan2) {
+  const c2 = cacheFor(scan2);
+  return c2.callerIndex ??= buildCallerIndex(scan2, importPairsFor(scan2));
+}
+function bm25DocsFor(scan2) {
+  const c2 = cacheFor(scan2);
+  return (c2.bm25 ??= { docs: buildDocs(scan2) }).docs;
+}
+function bm25TrigramsFor(scan2) {
+  const c2 = cacheFor(scan2);
+  const bm25 = c2.bm25 ??= { docs: buildDocs(scan2) };
+  return bm25.trigrams ??= buildTrigramIndex(bm25.docs);
+}
+function fileComplexityFor(scan2) {
+  const c2 = cacheFor(scan2);
+  if (!c2.fileComplexity) {
+    const m = /* @__PURE__ */ new Map();
+    for (const f of scan2.files) {
+      if (f.kind !== "code") continue;
+      m.set(f.rel, complexityOfSource(readText(join7(scan2.root, f.rel))));
+    }
+    c2.fileComplexity = m;
+  }
+  return c2.fileComplexity;
+}
+var caches;
+var init_derived = __esm({
+  "src/derived.ts"() {
+    "use strict";
+    init_resolve();
     init_graph();
+    init_symbols_json();
+    init_callers();
+    init_bm25();
+    init_complexity();
+    init_walk();
+    caches = /* @__PURE__ */ new WeakMap();
+  }
+});
+
+// src/callers.ts
+function computeImportPairs(scan2) {
+  return new Set(importPairsFor(scan2));
+}
+function buildCallerIndex(scan2, importPairs, opts = {}) {
+  const pairs = importPairs ?? importPairsFor(scan2);
+  const recall = opts.recall === true;
+  const defs = /* @__PURE__ */ new Map();
+  for (const f of scan2.files) {
+    const seen = /* @__PURE__ */ new Set();
+    for (const s of f.symbols) {
+      if (!s.exported || REFERENCE_KINDS3.has(s.kind)) continue;
+      if (seen.has(s.name)) continue;
+      seen.add(s.name);
+      let arr = defs.get(s.name);
+      if (!arr) defs.set(s.name, arr = []);
+      arr.push(s);
+    }
+  }
+  const localDefs = /* @__PURE__ */ new Map();
+  for (const f of scan2.files) {
+    const byName = /* @__PURE__ */ new Map();
+    for (const s of f.symbols) {
+      if (!REFERENCE_KINDS3.has(s.kind) && !byName.has(s.name)) byName.set(s.name, s);
+    }
+    localDefs.set(f.rel, byName);
+  }
+  const sites = /* @__PURE__ */ new Map();
+  const record = (def, caller) => {
+    let entry = sites.get(def.name + "\0" + def.file);
+    if (!entry) sites.set(def.name + "\0" + def.file, entry = { def, callers: [] });
+    entry.callers.push(caller);
+  };
+  for (const f of scan2.files) {
+    if (!f.calls?.length) continue;
+    const family = familyOf(f.lang);
+    const own = localDefs.get(f.rel);
+    for (const c2 of f.calls) {
+      const local = own.get(c2.name);
+      if (local) {
+        if (local.line !== c2.line)
+          record(local, recall ? { file: f.rel, line: c2.line, confidence: "corroborated" } : { file: f.rel, line: c2.line });
+        continue;
+      }
+      const cands = (defs.get(c2.name) ?? []).filter((d) => familyOf(d.lang) === family && d.file !== f.rel).map((d) => ({ file: d.file, lang: d.lang }));
+      if (!cands.length) continue;
+      const imported = cands.filter((d) => pairs.has(`${f.rel}|${d.file}`));
+      const chosen = family === "js" ? imported.length ? pickCandidate(f.rel, imported) : (
+        // JS/TS gate: no corroborating import → no binding. Recall mode
+        // relaxes this to a unique-repo-wide name match (issue #7).
+        recall && cands.length === 1 ? cands[0] : void 0
+      ) : imported.length ? pickCandidate(f.rel, imported) : pickCandidate(f.rel, cands);
+      if (!chosen) continue;
+      const def = defs.get(c2.name).find((d) => d.file === chosen.file);
+      record(
+        def,
+        recall ? { file: f.rel, line: c2.line, confidence: imported.length ? "corroborated" : "unique-name" } : { file: f.rel, line: c2.line }
+      );
+    }
+  }
+  const index = /* @__PURE__ */ new Map();
+  const keys = [...sites.keys()].sort(byStr);
+  for (const key of keys) {
+    const { def, callers } = sites.get(key);
+    callers.sort((a, b) => byStr(a.file, b.file) || a.line - b.line);
+    if (!index.has(def.name)) index.set(def.name, { def, callers });
+    else index.set(`${def.name}@${def.file}`, { def, callers });
+  }
+  return index;
+}
+function enclosingSymbol(scan2, file, line) {
+  const f = scan2.files.find((x) => x.rel === file);
+  if (!f?.symbols.length) return void 0;
+  return enclosingAmong(f.symbols, line);
+}
+function enclosingAmong(symbols, line) {
+  let best;
+  for (const s of symbols) {
+    if (REFERENCE_KINDS3.has(s.kind)) continue;
+    if (s.line > line) continue;
+    if (s.endLine !== void 0 && line > s.endLine) continue;
+    if (!best || s.line > best.line || s.line === best.line && (s.endLine ?? Infinity) <= (best.endLine ?? Infinity)) {
+      best = s;
+    }
+  }
+  return best;
+}
+function buildRawCallerIndex(scan2) {
+  const byName = /* @__PURE__ */ new Map();
+  for (const f of scan2.files) {
+    if (!f.calls?.length) continue;
+    const symbols = f.symbols.filter((s) => !REFERENCE_KINDS3.has(s.kind));
+    for (const c2 of f.calls) {
+      const site = { file: f.rel, line: c2.line };
+      if (c2.receiver !== void 0) site.receiver = c2.receiver;
+      const enc = enclosingAmong(symbols, c2.line);
+      if (enc) site.enclosingSymbol = enc;
+      let arr = byName.get(c2.name);
+      if (!arr) byName.set(c2.name, arr = []);
+      arr.push(site);
+    }
+  }
+  const index = /* @__PURE__ */ new Map();
+  for (const name2 of [...byName.keys()].sort(byStr)) {
+    const sites = byName.get(name2);
+    sites.sort((a, b) => byStr(a.file, b.file) || a.line - b.line);
+    index.set(name2, sites);
+  }
+  return index;
+}
+var REFERENCE_KINDS3;
+var init_callers = __esm({
+  "src/callers.ts"() {
+    "use strict";
+    init_calls();
+    init_derived();
+    init_sort();
+    REFERENCE_KINDS3 = /* @__PURE__ */ new Set(["reexport", "reexport-all", "default"]);
+  }
+});
+
+// src/query.ts
+import { join as join8 } from "path";
+function symbolsOverview(scan2, rel) {
+  const f = scan2.files.find((x) => x.rel === rel);
+  if (!f) return [];
+  return [...f.symbols].filter((s) => !REFERENCE_KINDS4.has(s.kind)).sort((a, b) => a.line - b.line || byStr(a.name, b.name));
+}
+function findSymbol(scan2, namePath, opts = {}) {
+  const segments = namePath.split("/").filter(Boolean);
+  if (!segments.length) return [];
+  const leaf = segments[segments.length - 1];
+  const parents = segments.slice(0, -1);
+  const matchName = (name2, wanted) => opts.substring ? name2.toLowerCase().includes(wanted.toLowerCase()) : name2 === wanted;
+  const out2 = [];
+  for (const f of scan2.files) {
+    for (const s of f.symbols) {
+      if (REFERENCE_KINDS4.has(s.kind)) continue;
+      if (!matchName(s.name, leaf)) continue;
+      if (parents.length) {
+        const parent = parents[parents.length - 1];
+        if (!s.parent || s.parent !== parent) continue;
+      }
+      out2.push({ ...s });
+    }
+  }
+  out2.sort(
+    (a, b) => Number(b.name === leaf) - Number(a.name === leaf) || byStr(a.file, b.file) || a.line - b.line
+  );
+  const capped = out2.slice(0, opts.maxResults ?? 50);
+  if (opts.includeBody) {
+    for (const m of capped) {
+      const end = m.endLine ?? m.line;
+      const content = readText(join8(scan2.root, m.file));
+      if (!content) continue;
+      m.body = content.split("\n").slice(m.line - 1, end).join("\n");
+    }
+  }
+  return capped;
+}
+function findReferences(scan2, name2) {
+  const defs = [];
+  for (const f of scan2.files) {
+    for (const s of f.symbols) {
+      if (s.name === name2 && !REFERENCE_KINDS4.has(s.kind)) defs.push(s);
+    }
+  }
+  defs.sort((a, b) => byStr(a.file, b.file) || a.line - b.line);
+  const index = callerIndexFor(scan2);
+  const entry = index.get(name2);
+  const callSites = entry ? [...entry.callers] : [];
+  const referencingFiles = /* @__PURE__ */ new Set();
+  const unique = uniqueDefsFor(scan2);
+  const defFile = unique.get(name2);
+  for (const f of scan2.files) {
+    if (f.rel === defFile) continue;
+    if (f.kind === "code" && f.idents?.includes(name2)) referencingFiles.add(f.rel);
+    else if (f.kind === "doc") {
+      const content = scan2.docText.get(f.rel);
+      if (content && new RegExp(`\\b${name2.replace(/[.*+?^${}()|[\]\\]/g, "\\$&")}\\b`).test(content)) {
+        referencingFiles.add(f.rel);
+      }
+    }
+  }
+  for (const site of callSites) referencingFiles.add(site.file);
+  return { defs, callSites, referencingFiles: [...referencingFiles].sort(byStr) };
+}
+var REFERENCE_KINDS4;
+var init_query = __esm({
+  "src/query.ts"() {
+    "use strict";
+    init_walk();
+    init_derived();
     init_sort();
     REFERENCE_KINDS4 = /* @__PURE__ */ new Set(["reexport", "reexport-all", "default"]);
   }
 });
 
 // src/edit.ts
-import { readFileSync as readFileSync3, writeFileSync } from "fs";
-import { join as join6 } from "path";
+import { readFileSync as readFileSync3, writeFileSync as writeFileSync2 } from "fs";
+import { join as join9 } from "path";
 function resolveUniqueSymbol(scan2, namePath, file) {
   let matches = findSymbol(scan2, namePath);
   if (file) matches = matches.filter((m) => m.file === file);
@@ -7878,15 +8278,15 @@ function readLines(abs) {
 function replaceSymbolBody(scan2, namePath, body2, file) {
   const sym = resolveUniqueSymbol(scan2, namePath, file);
   const end = sym.endLine ?? sym.line;
-  const abs = join6(scan2.root, sym.file);
+  const abs = join9(scan2.root, sym.file);
   const lines = readLines(abs);
   const newLines = body2.replace(/^\n+|\n+$/g, "").split("\n");
   lines.splice(sym.line - 1, end - sym.line + 1, ...newLines);
-  writeFileSync(abs, lines.join("\n"));
+  writeFileSync2(abs, lines.join("\n"));
   return { file: sym.file, startLine: sym.line, endLine: sym.line + newLines.length - 1, lines: newLines.length };
 }
 function insertAt(scan2, sym, body2, index, blankBefore, blankAfter) {
-  const abs = join6(scan2.root, sym.file);
+  const abs = join9(scan2.root, sym.file);
   const lines = readLines(abs);
   const minGap = SEPARATED_KINDS.has(sym.kind) ? 1 : 0;
   const newLines = body2.replace(/^\n+|\n+$/g, "").split("\n");
@@ -7895,7 +8295,7 @@ function insertAt(scan2, sym, body2, index, blankBefore, blankAfter) {
   block.push(...newLines);
   if (blankAfter && minGap && lines[index]?.trim() !== "") block.push("");
   lines.splice(index, 0, ...block);
-  writeFileSync(abs, lines.join("\n"));
+  writeFileSync2(abs, lines.join("\n"));
   return { file: sym.file, startLine: index + 1, endLine: index + block.length, lines: block.length };
 }
 function insertAfterSymbol(scan2, namePath, body2, file) {
@@ -7917,8 +8317,8 @@ var init_edit = __esm({
 });
 
 // src/memory.ts
-import { mkdirSync, readdirSync as readdirSync2, readFileSync as readFileSync4, rmSync, statSync as statSync2, writeFileSync as writeFileSync2 } from "fs";
-import { dirname as dirname2, join as join7 } from "path";
+import { mkdirSync as mkdirSync2, readdirSync as readdirSync2, readFileSync as readFileSync4, rmSync, statSync as statSync2, writeFileSync as writeFileSync3 } from "fs";
+import { dirname as dirname3, join as join10 } from "path";
 function sanitize(name2) {
   const clean = name2.replace(/^mem:/, "").replace(/\.md$/, "");
   if (!clean) throw new Error("memory name is empty");
@@ -7932,12 +8332,12 @@ function sanitize(name2) {
   return clean;
 }
 function memoryPath(repo, name2) {
-  return join7(repo, ...MEMORY_DIR, `${sanitize(name2)}.md`);
+  return join10(repo, ...MEMORY_DIR, `${sanitize(name2)}.md`);
 }
 function writeMemory(repo, name2, content) {
   const path = memoryPath(repo, name2);
-  mkdirSync(dirname2(path), { recursive: true });
-  writeFileSync2(path, content.endsWith("\n") ? content : content + "\n");
+  mkdirSync2(dirname3(path), { recursive: true });
+  writeFileSync3(path, content.endsWith("\n") ? content : content + "\n");
   return sanitize(name2);
 }
 function readMemory(repo, name2) {
@@ -7958,7 +8358,7 @@ function deleteMemory(repo, name2) {
   return true;
 }
 function listMemories(repo) {
-  const root = join7(repo, ...MEMORY_DIR);
+  const root = join10(repo, ...MEMORY_DIR);
   const out2 = [];
   const walk2 = (dir, prefix) => {
     let entries;
@@ -7968,7 +8368,7 @@ function listMemories(repo) {
       return;
     }
     for (const e of entries) {
-      if (e.isDirectory()) walk2(join7(dir, e.name), prefix ? `${prefix}/${e.name}` : e.name);
+      if (e.isDirectory()) walk2(join10(dir, e.name), prefix ? `${prefix}/${e.name}` : e.name);
       else if (e.name.endsWith(".md")) out2.push(prefix ? `${prefix}/${e.name.slice(0, -3)}` : e.name.slice(0, -3));
     }
   };
@@ -7985,7 +8385,7 @@ var init_memory = __esm({
 
 // src/workspaces.ts
 import { existsSync as existsSync2, readdirSync as readdirSync3, statSync as statSync3 } from "fs";
-import { join as join8 } from "path";
+import { join as join11 } from "path";
 function readJson(path, label, warnings) {
   const raw = readText(path);
   if (!raw) return void 0;
@@ -8036,7 +8436,7 @@ function wsGlobToRegExp(pat) {
   return new RegExp(`^${re}($|/)`);
 }
 function probeNodePkg(root, dir, kind, warnings) {
-  const path = join8(root, dir, "package.json");
+  const path = join11(root, dir, "package.json");
   if (!existsSync2(path)) return void 0;
   const manifest = `${dir}/package.json`;
   const pkg = readJson(path, manifest, warnings);
@@ -8050,7 +8450,7 @@ function probeNodePkg(root, dir, kind, warnings) {
   return out2;
 }
 function probeCargo(root, dir) {
-  const path = join8(root, dir, "Cargo.toml");
+  const path = join11(root, dir, "Cargo.toml");
   if (!existsSync2(path)) return void 0;
   const body2 = tomlSectionBody(readText(path), "package");
   const out2 = {
@@ -8064,18 +8464,18 @@ function probeCargo(root, dir) {
   return out2;
 }
 function probeGoMod(root, dir) {
-  const path = join8(root, dir, "go.mod");
+  const path = join11(root, dir, "go.mod");
   if (!existsSync2(path)) return void 0;
   const name2 = readText(path).match(/^module\s+(\S+)/m)?.[1] ?? dir;
   return { name: name2, dir, kind: "go", manifest: `${dir}/go.mod` };
 }
 function probeMaven(root, dir) {
-  const path = join8(root, dir, "pom.xml");
+  const path = join11(root, dir, "pom.xml");
   if (!existsSync2(path)) return void 0;
   return { name: ownArtifactId(readText(path)) ?? dir, dir, kind: "maven", manifest: `${dir}/pom.xml` };
 }
 function probePyproject(root, dir) {
-  const path = join8(root, dir, "pyproject.toml");
+  const path = join11(root, dir, "pyproject.toml");
   if (!existsSync2(path)) return void 0;
   const toml = readText(path);
   const project = tomlSectionBody(toml, "project");
@@ -8091,7 +8491,7 @@ function probePyproject(root, dir) {
   return out2;
 }
 function probeComposer(root, dir, warnings) {
-  const path = join8(root, dir, "composer.json");
+  const path = join11(root, dir, "composer.json");
   if (!existsSync2(path)) return void 0;
   const manifest = `${dir}/composer.json`;
   const pkg = readJson(path, manifest, warnings);
@@ -8105,7 +8505,7 @@ function probeComposer(root, dir, warnings) {
   return out2;
 }
 function probeNxProject(root, dir, warnings) {
-  const path = join8(root, dir, "project.json");
+  const path = join11(root, dir, "project.json");
   if (!existsSync2(path)) return void 0;
   const manifest = `${dir}/project.json`;
   const proj = readJson(path, manifest, warnings);
@@ -8118,7 +8518,7 @@ function probeNxProject(root, dir, warnings) {
 }
 function probeGradle(root, dir) {
   for (const f of ["build.gradle", "build.gradle.kts"]) {
-    if (existsSync2(join8(root, dir, f))) {
+    if (existsSync2(join11(root, dir, f))) {
       return { name: dir, dir, kind: "gradle", manifest: `${dir}/${f}` };
     }
   }
@@ -8153,7 +8553,7 @@ function addPackage(root, dir, found, kind, warnings) {
 }
 function isDirAt(root, rel) {
   try {
-    return statSync3(join8(root, rel)).isDirectory();
+    return statSync3(join11(root, rel)).isDirectory();
   } catch {
     return false;
   }
@@ -8161,7 +8561,7 @@ function isDirAt(root, rel) {
 function subdirsOf(root, base) {
   let entries;
   try {
-    entries = readdirSync3(base ? join8(root, base) : root, { withFileTypes: true });
+    entries = readdirSync3(base ? join11(root, base) : root, { withFileTypes: true });
   } catch {
     return [];
   }
@@ -8223,14 +8623,14 @@ function npmFamilyPatterns(root, warnings) {
     if (t.startsWith("!")) negations.push(t.slice(1));
     else positives.push({ pattern: t, kind });
   };
-  const pkg = readJson(join8(root, "package.json"), "package.json", warnings);
+  const pkg = readJson(join11(root, "package.json"), "package.json", warnings);
   const ws = pkg?.workspaces;
   if (Array.isArray(ws)) {
     for (const x of ws) if (typeof x === "string") push(x, "npm");
   } else if (ws && typeof ws === "object" && Array.isArray(ws.packages)) {
     for (const x of ws.packages) if (typeof x === "string") push(x, "npm");
   }
-  const pnpm = readText(join8(root, "pnpm-workspace.yaml"));
+  const pnpm = readText(join11(root, "pnpm-workspace.yaml"));
   let inPackages = false;
   for (const line of pnpm.split(/\r?\n/)) {
     if (/^\S/.test(line)) {
@@ -8244,11 +8644,11 @@ function npmFamilyPatterns(root, warnings) {
   return { positives, negations };
 }
 function fallbackNpmPatterns(root, warnings) {
-  const lerna = readJson(join8(root, "lerna.json"), "lerna.json", warnings);
+  const lerna = readJson(join11(root, "lerna.json"), "lerna.json", warnings);
   if (lerna && Array.isArray(lerna.packages)) {
     return lerna.packages.filter((x) => typeof x === "string").map((pattern) => ({ pattern, kind: "lerna" }));
   }
-  const nx = readJson(join8(root, "nx.json"), "nx.json", warnings);
+  const nx = readJson(join11(root, "nx.json"), "nx.json", warnings);
   if (nx) {
     const layout = nx.workspaceLayout ?? {};
     const appsDir = typeof layout.appsDir === "string" ? layout.appsDir : "apps";
@@ -8258,7 +8658,7 @@ function fallbackNpmPatterns(root, warnings) {
   return [];
 }
 function detectCargoMembers(root, found, warnings) {
-  const toml = readText(join8(root, "Cargo.toml"));
+  const toml = readText(join11(root, "Cargo.toml"));
   if (!toml) return;
   const body2 = tomlSectionBody(toml, "workspace");
   if (!body2) return;
@@ -8273,7 +8673,7 @@ function detectCargoMembers(root, found, warnings) {
   }
 }
 function detectGoWork(root, found, warnings) {
-  const gowork = readText(join8(root, "go.work"));
+  const gowork = readText(join11(root, "go.work"));
   if (!gowork) return;
   const dirs = [];
   for (const block of gowork.matchAll(/^use\s*\(([\s\S]*?)\)/gm)) {
@@ -8289,7 +8689,7 @@ function detectGoWork(root, found, warnings) {
   }
 }
 function detectMavenModules(root, found, warnings) {
-  const pom = readText(join8(root, "pom.xml"));
+  const pom = readText(join11(root, "pom.xml"));
   if (!pom) return;
   const modules = pom.match(/<modules>([\s\S]*?)<\/modules>/)?.[1];
   if (!modules) return;
@@ -8298,7 +8698,7 @@ function detectMavenModules(root, found, warnings) {
   }
 }
 function detectUvMembers(root, found, warnings) {
-  const toml = readText(join8(root, "pyproject.toml"));
+  const toml = readText(join11(root, "pyproject.toml"));
   if (!toml) return;
   const body2 = tomlSectionBody(toml, "tool.uv.workspace");
   if (!body2) return;
@@ -8313,7 +8713,7 @@ function detectUvMembers(root, found, warnings) {
   }
 }
 function detectComposerPathRepos(root, found, warnings) {
-  const composer = readJson(join8(root, "composer.json"), "composer.json", warnings);
+  const composer = readJson(join11(root, "composer.json"), "composer.json", warnings);
   const repos = composer?.repositories;
   if (!Array.isArray(repos)) return;
   for (const r of repos) {
@@ -8324,7 +8724,7 @@ function detectComposerPathRepos(root, found, warnings) {
 }
 function detectGradleIncludes(root, found, warnings) {
   for (const f of ["settings.gradle", "settings.gradle.kts"]) {
-    const text = readText(join8(root, f));
+    const text = readText(join11(root, f));
     if (!text) continue;
     for (const line of text.split(/\r?\n/)) {
       if (!/^\s*include[\s(]/.test(line)) continue;
@@ -8336,7 +8736,7 @@ function detectGradleIncludes(root, found, warnings) {
   }
 }
 function npmEdges(root, pkg, byName, warnings) {
-  const manifest = readJson(join8(root, pkg.dir, "package.json"), `${pkg.dir}/package.json`, warnings);
+  const manifest = readJson(join11(root, pkg.dir, "package.json"), `${pkg.dir}/package.json`, warnings);
   if (!manifest) return [];
   const edges = /* @__PURE__ */ new Set();
   for (const field of ["dependencies", "devDependencies", "peerDependencies"]) {
@@ -8359,7 +8759,7 @@ function normalizeDepPath(fromDir, rel) {
   return out2.join("/");
 }
 function cargoEdges(root, pkg, byName, byDir) {
-  const toml = readText(join8(root, pkg.dir, "Cargo.toml"));
+  const toml = readText(join11(root, pkg.dir, "Cargo.toml"));
   if (!toml) return [];
   const edges = /* @__PURE__ */ new Set();
   for (const section of ["dependencies", "dev-dependencies", "build-dependencies"]) {
@@ -8383,7 +8783,7 @@ function cargoEdges(root, pkg, byName, byDir) {
   return [...edges];
 }
 function goPkgEdges(root, pkg, byName, byDir) {
-  const gomod = readText(join8(root, pkg.dir, "go.mod"));
+  const gomod = readText(join11(root, pkg.dir, "go.mod"));
   if (!gomod) return [];
   const edges = /* @__PURE__ */ new Set();
   for (const m of gomod.matchAll(/^\s*(?:require\s+)?([^\s/(][^\s]*)\s+v[^\s]+/gm)) {
@@ -8397,7 +8797,7 @@ function goPkgEdges(root, pkg, byName, byDir) {
   return [...edges];
 }
 function mavenEdges(root, pkg, byName) {
-  const pom = readText(join8(root, pkg.dir, "pom.xml"));
+  const pom = readText(join11(root, pkg.dir, "pom.xml"));
   if (!pom) return [];
   const edges = /* @__PURE__ */ new Set();
   for (const m of pom.replace(/<parent>[\s\S]*?<\/parent>/g, "").matchAll(/<dependency>([\s\S]*?)<\/dependency>/g)) {
@@ -8407,7 +8807,7 @@ function mavenEdges(root, pkg, byName) {
   return [...edges];
 }
 function uvEdges(root, pkg, byName) {
-  const toml = readText(join8(root, pkg.dir, "pyproject.toml"));
+  const toml = readText(join11(root, pkg.dir, "pyproject.toml"));
   if (!toml) return [];
   const edges = /* @__PURE__ */ new Set();
   const project = tomlSectionBody(toml, "project");
@@ -8427,7 +8827,7 @@ function uvEdges(root, pkg, byName) {
   return [...edges];
 }
 function composerEdges(root, pkg, byName, warnings) {
-  const manifest = readJson(join8(root, pkg.dir, "composer.json"), `${pkg.dir}/composer.json`, warnings);
+  const manifest = readJson(join11(root, pkg.dir, "composer.json"), `${pkg.dir}/composer.json`, warnings);
   if (!manifest) return [];
   const edges = /* @__PURE__ */ new Set();
   for (const field of ["require", "require-dev"]) {
@@ -8441,7 +8841,7 @@ function composerEdges(root, pkg, byName, warnings) {
 }
 function gradleEdges(root, pkg, byName, byDir) {
   for (const f of ["build.gradle", "build.gradle.kts"]) {
-    const text = readText(join8(root, pkg.dir, f));
+    const text = readText(join11(root, pkg.dir, f));
     if (!text) continue;
     const edges = /* @__PURE__ */ new Set();
     for (const m of text.matchAll(/project\s*\(\s*["']:?([^"']+)["']\s*\)/g)) {
@@ -8977,117 +9377,50 @@ var init_tests_map = __esm({
     ];
     TEST_DIR = /(^|\/)(tests?|__tests?__|spec|specs|e2e)(\/|$)/i;
   }
-});
-
-// src/surprise.ts
-function computeSurprises(graph) {
-  const commOf = /* @__PURE__ */ new Map();
-  const tierOf2 = /* @__PURE__ */ new Map();
-  for (const m of graph.modules) {
-    if (m.community !== void 0) commOf.set(m.slug, m.community);
-    tierOf2.set(m.slug, m.tier);
-  }
-  const pairCount = /* @__PURE__ */ new Map();
-  const pairKey = (a, b) => a < b ? `${a}:${b}` : `${b}:${a}`;
-  const candidates = [];
-  for (const e of graph.moduleEdges) {
-    if (e.dangling) continue;
-    const ca = commOf.get(e.from);
-    const cb = commOf.get(e.to);
-    if (ca === void 0 || cb === void 0 || ca === cb) continue;
-    pairCount.set(pairKey(ca, cb), (pairCount.get(pairKey(ca, cb)) ?? 0) + 1);
-    if (!DEP_KINDS.has(e.kind)) continue;
-    if (tierOf2.get(e.to) === 0) continue;
-    candidates.push({ edge: e, comms: [ca, cb] });
-  }
-  return candidates.filter((c2) => pairCount.get(pairKey(c2.comms[0], c2.comms[1])) <= MAX_PAIR_EDGES).map((c2) => ({
-    from: c2.edge.from,
-    to: c2.edge.to,
-    kind: c2.edge.kind,
-    weight: c2.edge.weight,
-    communities: c2.comms,
-    pairEdges: pairCount.get(pairKey(c2.comms[0], c2.comms[1]))
-  })).sort((a, b) => a.pairEdges - b.pairEdges || byStr(a.from, b.from) || byStr(a.to, b.to)).slice(0, SURPRISE_CAP);
-}
-function isSurprising(graph, from, to) {
-  const list = graph.surprises ?? computeSurprises(graph);
-  return list.some((s) => s.from === from && s.to === to);
-}
-var SURPRISE_CAP, MAX_PAIR_EDGES, DEP_KINDS;
-var init_surprise = __esm({
-  "src/surprise.ts"() {
-    "use strict";
-    init_sort();
-    SURPRISE_CAP = 24;
-    MAX_PAIR_EDGES = 2;
-    DEP_KINDS = /* @__PURE__ */ new Set(["import", "call", "use"]);
-  }
-});
-
-// src/render/symbols-json.ts
-function computeSymbolRefs(scan2) {
-  const unique = uniqueSymbolDefs(scan2);
-  const refs = /* @__PURE__ */ new Map();
-  if (!unique.size) return refs;
-  const add = (name2, file) => {
-    let set = refs.get(name2);
-    if (!set) refs.set(name2, set = /* @__PURE__ */ new Set());
-    set.add(file);
-  };
-  for (const f of scan2.files) {
-    if (f.kind === "code" && f.idents) {
-      for (const id of f.idents) {
-        const target = unique.get(id);
-        if (target && target !== f.rel) add(id, f.rel);
-      }
-    } else if (f.kind === "doc") {
-      const content = scan2.docText.get(f.rel);
-      if (!content) continue;
-      for (const tok of content.split(/[^A-Za-z0-9_]+/)) {
-        const target = unique.get(tok);
-        if (target && target !== f.rel) add(tok, f.rel);
-      }
-    }
-  }
-  return refs;
-}
-function buildSymbolIndex(scan2, refs = /* @__PURE__ */ new Map()) {
-  const defsByName = /* @__PURE__ */ new Map();
-  for (const f of scan2.files) {
-    for (const s of f.symbols) {
-      let arr = defsByName.get(s.name);
-      if (!arr) defsByName.set(s.name, arr = []);
-      arr.push({
-        file: s.file,
-        line: s.line,
-        ...s.endLine !== void 0 ? { endLine: s.endLine } : {},
-        kind: s.kind,
-        exported: s.exported,
-        lang: s.lang,
-        ...s.parent ? { parent: s.parent } : {}
-      });
-    }
-  }
-  const defs = {};
-  for (const name2 of [...defsByName.keys()].sort(byStr)) {
-    defs[name2] = defsByName.get(name2).slice().sort((a, b) => byStr(a.file, b.file) || a.line - b.line || byStr(a.kind, b.kind));
+});
+
+// src/surprise.ts
+function computeSurprises(graph) {
+  const commOf = /* @__PURE__ */ new Map();
+  const tierOf2 = /* @__PURE__ */ new Map();
+  for (const m of graph.modules) {
+    if (m.community !== void 0) commOf.set(m.slug, m.community);
+    tierOf2.set(m.slug, m.tier);
   }
-  const refsOut = {};
-  for (const name2 of [...refs.keys()].sort(byStr)) {
-    const files = [...refs.get(name2)].sort(byStr);
-    if (files.length) refsOut[name2] = files;
+  const pairCount = /* @__PURE__ */ new Map();
+  const pairKey = (a, b) => a < b ? `${a}:${b}` : `${b}:${a}`;
+  const candidates = [];
+  for (const e of graph.moduleEdges) {
+    if (e.dangling) continue;
+    const ca = commOf.get(e.from);
+    const cb = commOf.get(e.to);
+    if (ca === void 0 || cb === void 0 || ca === cb) continue;
+    pairCount.set(pairKey(ca, cb), (pairCount.get(pairKey(ca, cb)) ?? 0) + 1);
+    if (!DEP_KINDS.has(e.kind)) continue;
+    if (tierOf2.get(e.to) === 0) continue;
+    candidates.push({ edge: e, comms: [ca, cb] });
   }
-  return { schemaVersion: SCHEMA_VERSION, defs, refs: refsOut };
+  return candidates.filter((c2) => pairCount.get(pairKey(c2.comms[0], c2.comms[1])) <= MAX_PAIR_EDGES).map((c2) => ({
+    from: c2.edge.from,
+    to: c2.edge.to,
+    kind: c2.edge.kind,
+    weight: c2.edge.weight,
+    communities: c2.comms,
+    pairEdges: pairCount.get(pairKey(c2.comms[0], c2.comms[1]))
+  })).sort((a, b) => a.pairEdges - b.pairEdges || byStr(a.from, b.from) || byStr(a.to, b.to)).slice(0, SURPRISE_CAP);
 }
-function renderSymbolsJson(index) {
-  return JSON.stringify(index, null, 2) + "\n";
+function isSurprising(graph, from, to) {
+  const list = graph.surprises ?? computeSurprises(graph);
+  return list.some((s) => s.from === from && s.to === to);
 }
-var init_symbols_json = __esm({
-  "src/render/symbols-json.ts"() {
+var SURPRISE_CAP, MAX_PAIR_EDGES, DEP_KINDS;
+var init_surprise = __esm({
+  "src/surprise.ts"() {
     "use strict";
-    init_types();
     init_sort();
-    init_graph();
+    SURPRISE_CAP = 24;
+    MAX_PAIR_EDGES = 2;
+    DEP_KINDS = /* @__PURE__ */ new Set(["import", "call", "use"]);
   }
 });
 
@@ -9110,8 +9443,10 @@ var init_graph_json = __esm({
 
 // src/pipeline.ts
 function buildIndexArtifacts(repo, opts = {}) {
-  const scan2 = scanRepo(repo, opts);
-  const ctx = buildResolveContext(scan2);
+  return buildArtifactsFromScan(scanRepo(repo, opts), opts);
+}
+function buildArtifactsFromScan(scan2, opts = {}) {
+  const ctx = resolveContextFor(scan2);
   const { modules, moduleOf } = buildModules(scan2);
   const graph = buildGraph(scan2, ctx, modules, moduleOf, opts.meta);
   const communities = detectCommunities(graph.modules, graph.moduleEdges, opts.previousCommunities);
@@ -9130,14 +9465,14 @@ function buildIndexArtifacts(repo, opts = {}) {
   }
   const surprises = computeSurprises(graph);
   if (surprises.length) graph.surprises = surprises;
-  const symbols = buildSymbolIndex(scan2, computeSymbolRefs(scan2));
+  const symbols = buildSymbolIndex(scan2, symbolRefsFor(scan2));
   return { scan: scan2, graph, symbols };
 }
 var init_pipeline = __esm({
   "src/pipeline.ts"() {
     "use strict";
     init_scan();
-    init_resolve();
+    init_derived();
     init_modules();
     init_graph();
     init_community();
@@ -9159,275 +9494,99 @@ function rgBackend(root, pattern, opts) {
     "--null",
     // path\0line:text — a `:12:` inside a filename can't corrupt parsing
     "--color=never",
-    "--no-messages",
-    "--hidden",
-    "--no-require-git",
-    "--no-ignore-global",
-    "--no-ignore-exclude",
-    "--no-ignore-parent",
-    "--no-ignore-dot",
-    "--max-filesize",
-    "1M"
-  ];
-  for (const d of IGNORE_DIRS) args2.push("--glob", `!**/${d}/**`);
-  for (const l of LOCKFILES) args2.push("--iglob", `!**/${l}`);
-  for (const ext of BINARY_EXT) args2.push("--iglob", `!**/*${ext}`);
-  args2.push("--glob", "!*.min.js", "--glob", "!*.min.css");
-  if (opts.ignoreCase) args2.push("--ignore-case");
-  const user = opts.globs ?? [];
-  const anchor = (g) => g.startsWith("/") ? g : `/${g}`;
-  for (const g of user.filter((g2) => !g2.startsWith("!"))) args2.push("--glob", anchor(g));
-  for (const g of user.filter((g2) => g2.startsWith("!"))) args2.push("--glob", `!${anchor(g.slice(1))}`);
-  args2.push("--regexp", pattern, "./");
-  const res = sh("rg", args2, { cwd: root });
-  if (res.missing || !res.ok && res.status !== 1) return void 0;
-  const hits = [];
-  for (const line of res.stdout.split("\n")) {
-    if (!line) continue;
-    const nul = line.indexOf("\0");
-    if (nul === -1) continue;
-    const file = line.slice(0, nul).replace(/^\.\//, "");
-    const rest = line.slice(nul + 1);
-    const colon = rest.indexOf(":");
-    if (colon === -1) continue;
-    hits.push({ file, line: Number(rest.slice(0, colon)), text: rest.slice(colon + 1) });
-  }
-  return hits;
-}
-function jsBackend(root, re, opts) {
-  const filter = compileGlobFilter(opts.globs?.map((g) => g.replace(/^(!?)\//, "$1")));
-  const hits = [];
-  for (const f of walk(root).files) {
-    if (filter && !filter(f.rel)) continue;
-    const content = readText(f.abs);
-    if (!content) continue;
-    const lines = content.split("\n");
-    for (let i2 = 0; i2 < lines.length; i2++) {
-      if (re.test(lines[i2])) hits.push({ file: f.rel, line: i2 + 1, text: lines[i2] });
-    }
-  }
-  return hits;
-}
-function grepRepo(root, pattern, opts = {}) {
-  const re = new RegExp(pattern, opts.ignoreCase ? "i" : "");
-  const max = opts.maxHits ?? DEFAULT_MAX_HITS;
-  let hits;
-  if (!opts.noRipgrep && have("rg")) hits = rgBackend(root, pattern, opts);
-  hits ??= jsBackend(root, re, opts);
-  return sortHits(hits).slice(0, max);
-}
-var DEFAULT_MAX_HITS;
-var init_grep = __esm({
-  "src/grep.ts"() {
-    "use strict";
-    init_walk();
-    init_glob();
-    init_util();
-    init_sort();
-    DEFAULT_MAX_HITS = 200;
-  }
-});
-
-// src/bm25.ts
-function subtokens(raw) {
-  const folded = foldText(raw).replace(/([a-z0-9])([A-Z])/g, "$1 $2").replace(/([A-Z]+)([A-Z][a-z])/g, "$1 $2");
-  const out2 = [];
-  const seen = /* @__PURE__ */ new Set();
-  const push = (t) => {
-    if (t.length < 2 || seen.has(t)) return;
-    seen.add(t);
-    out2.push(t);
-  };
-  if (!/\s/.test(raw.trim())) push(foldText(raw).toLowerCase().replace(/[^a-z0-9_]+/g, ""));
-  for (const part of folded.split(/[^A-Za-z0-9]+/)) push(part.toLowerCase());
-  return out2;
-}
-function addTerms(doc, text) {
-  for (const t of subtokens(text)) {
-    doc.tf.set(t, (doc.tf.get(t) ?? 0) + 1);
-    doc.len++;
-  }
-}
-function buildDocs(scan2) {
-  const docs = [];
-  for (const f of scan2.files) {
-    const doc = { file: f.rel, tf: /* @__PURE__ */ new Map(), len: 0, symbols: [] };
-    const seenSym = /* @__PURE__ */ new Set();
-    for (const s of f.symbols) {
-      addTerms(doc, s.name);
-      if (!seenSym.has(s.name)) {
-        seenSym.add(s.name);
-        doc.symbols.push(s.name);
-      }
-    }
-    for (const seg of f.rel.split("/")) addTerms(doc, seg);
-    for (const h of f.headings) addTerms(doc, h);
-    if (f.summary) addTerms(doc, f.summary);
-    docs.push(doc);
-  }
-  return docs;
-}
-function charTrigrams(term) {
-  const padded = `^^${term}$$`;
-  const grams = /* @__PURE__ */ new Set();
-  for (let i2 = 0; i2 + 3 <= padded.length; i2++) grams.add(padded.slice(i2, i2 + 3));
-  return grams;
-}
-function diceCoefficient(a, b) {
-  if (!a.size || !b.size) return 0;
-  let inter = 0;
-  for (const g of a) if (b.has(g)) inter++;
-  return 2 * inter / (a.size + b.size);
-}
-function buildTrigramIndex(docs) {
-  const index = /* @__PURE__ */ new Map();
-  for (const d of docs) {
-    for (const term of d.tf.keys()) {
-      if (!index.has(term)) index.set(term, charTrigrams(term));
-    }
-  }
-  return index;
-}
-function searchIndex(scan2, query, opts = {}) {
-  const terms = [];
-  const seen = /* @__PURE__ */ new Set();
-  for (const kw of keywords(query)) {
-    for (const t of subtokens(kw)) {
-      if (seen.has(t)) continue;
-      seen.add(t);
-      terms.push(t);
-    }
-  }
-  if (!terms.length) return [];
-  const docs = buildDocs(scan2);
-  const n = docs.length;
-  if (!n) return [];
-  let totalLen = 0;
-  for (const d of docs) totalLen += d.len;
-  const avgLen = totalLen / n || 1;
-  const df = /* @__PURE__ */ new Map();
-  for (const t of terms) {
-    let count = 0;
-    for (const d of docs) if (d.tf.has(t)) count++;
-    df.set(t, count);
-  }
-  const fuzzyEnabled = opts.fuzzy ?? true;
-  const fuzzyCandidates = /* @__PURE__ */ new Map();
-  if (fuzzyEnabled) {
-    const unmatched = terms.filter((t) => df.get(t) === 0);
-    if (unmatched.length) {
-      const trigramIndex = buildTrigramIndex(docs);
-      for (const t of unmatched) {
-        const grams = charTrigrams(t);
-        const candidates = [];
-        for (const [vocabTerm, vocabGrams] of trigramIndex) {
-          const dice = diceCoefficient(grams, vocabGrams);
-          if (dice >= FUZZY_DICE_THRESHOLD) candidates.push({ term: vocabTerm, dice });
-        }
-        candidates.sort((a, b) => b.dice - a.dice || byStr(a.term, b.term));
-        fuzzyCandidates.set(t, candidates.slice(0, FUZZY_CAP));
-      }
-    }
-  }
-  const vocabDf = /* @__PURE__ */ new Map();
-  const dfOfVocabTerm = (term) => {
-    const known = df.get(term) ?? vocabDf.get(term);
-    if (known !== void 0) return known;
-    let count = 0;
-    for (const d of docs) if (d.tf.has(term)) count++;
-    vocabDf.set(term, count);
-    return count;
-  };
-  const results = [];
-  for (const d of docs) {
-    let score = 0;
-    const matched = [];
-    const symbolTerms = /* @__PURE__ */ new Set();
-    const fuzzyHit = /* @__PURE__ */ new Set();
-    for (const t of terms) {
-      const tf = d.tf.get(t);
-      if (tf) {
-        matched.push(t);
-        symbolTerms.add(t);
-        const idf = Math.log(1 + (n - df.get(t) + 0.5) / (df.get(t) + 0.5));
-        score += idf * (tf * (K1 + 1)) / (tf + K1 * (1 - B + B * d.len / avgLen));
-        continue;
-      }
-      const candidates = fuzzyCandidates.get(t);
-      if (!candidates) continue;
-      for (const cand of candidates) {
-        const ctf = d.tf.get(cand.term);
-        if (!ctf) continue;
-        const cdf = dfOfVocabTerm(cand.term);
-        const idf = Math.log(1 + (n - cdf + 0.5) / (cdf + 0.5));
-        const contribution = idf * (ctf * (K1 + 1)) / (ctf + K1 * (1 - B + B * d.len / avgLen));
-        score += contribution * cand.dice;
-        symbolTerms.add(cand.term);
-        fuzzyHit.add(t);
-      }
+    "--no-messages",
+    "--hidden",
+    "--no-require-git",
+    "--no-ignore-global",
+    "--no-ignore-exclude",
+    "--no-ignore-parent",
+    "--no-ignore-dot",
+    "--max-filesize",
+    "1M"
+  ];
+  for (const d of IGNORE_DIRS) args2.push("--glob", `!**/${d}/**`);
+  for (const l of LOCKFILES) args2.push("--iglob", `!**/${l}`);
+  for (const ext of BINARY_EXT) args2.push("--iglob", `!**/*${ext}`);
+  args2.push("--glob", "!*.min.js", "--glob", "!*.min.css");
+  if (opts.ignoreCase) args2.push("--ignore-case");
+  const user = opts.globs ?? [];
+  const anchor = (g) => g.startsWith("/") ? g : `/${g}`;
+  for (const g of user.filter((g2) => !g2.startsWith("!"))) args2.push("--glob", anchor(g));
+  for (const g of user.filter((g2) => g2.startsWith("!"))) args2.push("--glob", `!${anchor(g.slice(1))}`);
+  args2.push("--regexp", pattern, "./");
+  const res = sh("rg", args2, { cwd: root });
+  if (res.missing || !res.ok && res.status !== 1) return void 0;
+  const hits = [];
+  for (const line of res.stdout.split("\n")) {
+    if (!line) continue;
+    const nul = line.indexOf("\0");
+    if (nul === -1) continue;
+    const file = line.slice(0, nul).replace(/^\.\//, "");
+    const rest = line.slice(nul + 1);
+    const colon = rest.indexOf(":");
+    if (colon === -1) continue;
+    hits.push({ file, line: Number(rest.slice(0, colon)), text: rest.slice(colon + 1) });
+  }
+  return hits;
+}
+function jsBackend(root, re, opts) {
+  const filter = compileGlobFilter(opts.globs?.map((g) => g.replace(/^(!?)\//, "$1")));
+  const hits = [];
+  for (const f of walk(root).files) {
+    if (filter && !filter(f.rel)) continue;
+    const content = readText(f.abs);
+    if (!content) continue;
+    const lines = content.split("\n");
+    for (let i2 = 0; i2 < lines.length; i2++) {
+      if (re.test(lines[i2])) hits.push({ file: f.rel, line: i2 + 1, text: lines[i2] });
     }
-    if (!matched.length && !fuzzyHit.size) continue;
-    const scored = d.symbols.map((name2) => {
-      const toks = new Set(subtokens(name2));
-      let hits = 0;
-      for (const t of symbolTerms) if (toks.has(t)) hits++;
-      return { name: name2, hits };
-    }).filter((s) => s.hits > 0).sort((a, b) => b.hits - a.hits || byStr(a.name, b.name));
-    const result = {
-      file: d.file,
-      score: Number(score.toFixed(4)),
-      matchedTerms: matched.sort(byStr),
-      topSymbols: scored.slice(0, TOP_SYMBOLS).map((s) => s.name)
-    };
-    if (fuzzyHit.size) result.fuzzyTerms = [...fuzzyHit].sort(byStr);
-    results.push(result);
   }
-  results.sort((a, b) => b.score - a.score || byStr(a.file, b.file));
-  return results.slice(0, opts.limit ?? DEFAULT_LIMIT);
+  return hits;
 }
-var K1, B, DEFAULT_LIMIT, TOP_SYMBOLS, FUZZY_DICE_THRESHOLD, FUZZY_CAP;
-var init_bm25 = __esm({
-  "src/bm25.ts"() {
+function grepRepo(root, pattern, opts = {}) {
+  const re = new RegExp(pattern, opts.ignoreCase ? "i" : "");
+  const max = opts.maxHits ?? DEFAULT_MAX_HITS;
+  let hits;
+  if (!opts.noRipgrep && have("rg")) hits = rgBackend(root, pattern, opts);
+  hits ??= jsBackend(root, re, opts);
+  return sortHits(hits).slice(0, max);
+}
+var DEFAULT_MAX_HITS;
+var init_grep = __esm({
+  "src/grep.ts"() {
     "use strict";
+    init_walk();
+    init_glob();
     init_util();
     init_sort();
-    K1 = 1.2;
-    B = 0.75;
-    DEFAULT_LIMIT = 20;
-    TOP_SYMBOLS = 5;
-    FUZZY_DICE_THRESHOLD = 0.6;
-    FUZZY_CAP = 3;
+    DEFAULT_MAX_HITS = 200;
   }
 });
 
 // src/embed/model.ts
-import { createHash as createHash2 } from "crypto";
+import { createHash as createHash3 } from "crypto";
 import { existsSync as existsSync3, readFileSync as readFileSync5 } from "fs";
-import { join as join10 } from "path";
+import { join as join13 } from "path";
 function resolveEmbedModelDir(repo) {
   const env = process.env.CODEINDEX_EMBED_DIR;
   const candidates = [];
   if (env) candidates.push(env);
-  if (repo) candidates.push(join10(repo, ".codeindex", DEFAULT_EMBED_DIRNAME));
-  candidates.push(join10(process.cwd(), ".codeindex", DEFAULT_EMBED_DIRNAME));
+  if (repo) candidates.push(join13(repo, ".codeindex", DEFAULT_EMBED_DIRNAME));
+  candidates.push(join13(process.cwd(), ".codeindex", DEFAULT_EMBED_DIRNAME));
   for (const c2 of candidates) {
-    if (existsSync3(join10(c2, "model.json"))) return c2;
+    if (existsSync3(join13(c2, "model.json"))) return c2;
   }
   return void 0;
 }
 function hasEmbedModel(repo) {
   return resolveEmbedModelDir(repo) !== void 0;
 }
-function loadEmbedModel(dir) {
-  if (!dir) return void 0;
-  const path = join10(dir, "model.json");
-  if (!existsSync3(path)) return void 0;
-  const raw = JSON.parse(readFileSync5(path, "utf8"));
-  const { modelId, dim, vocab, weights } = raw;
-  if (typeof modelId !== "string" || !modelId) throw new Error(`embed model: missing modelId in ${path}`);
-  if (!Number.isInteger(dim) || dim <= 0) throw new Error(`embed model: bad dim ${dim} in ${path}`);
+function parseEmbedModel(raw, source) {
+  const { modelId, dim, vocab, weights, unk: rawUnk } = raw ?? {};
+  if (typeof modelId !== "string" || !modelId) throw new Error(`embed model: missing modelId in ${source}`);
+  if (!Number.isInteger(dim) || dim <= 0) throw new Error(`embed model: bad dim ${dim} in ${source}`);
   if (!Array.isArray(vocab) || !Array.isArray(weights) || vocab.length !== weights.length) {
-    throw new Error(`embed model: vocab/weights length mismatch in ${path}`);
+    throw new Error(`embed model: vocab/weights length mismatch in ${source}`);
   }
   const vocabSize = vocab.length;
   const flat = new Float64Array(vocabSize * dim);
@@ -9442,10 +9601,17 @@ function loadEmbedModel(dir) {
     }
     for (let d = 0; d < dim; d++) flat[i2 * dim + d] = Number(row[d]);
   }
-  const unk = typeof raw.unk === "string" ? raw.unk : "[UNK]";
+  const unk = typeof rawUnk === "string" ? rawUnk : "[UNK]";
   const unkId = vmap.has(unk) ? vmap.get(unk) : -1;
   return { modelId, dim, unk, unkId, vocabSize, vocab: vmap, weights: flat };
 }
+function loadEmbedModel(dir) {
+  if (!dir) return void 0;
+  const path = join13(dir, "model.json");
+  if (!existsSync3(path)) return void 0;
+  const raw = JSON.parse(readFileSync5(path, "utf8"));
+  return parseEmbedModel(raw, path);
+}
 function resolveEmbedPullUrl() {
   const env = process.env.CODEINDEX_EMBED_URL;
   if (env && env.trim()) return { url: env.trim() };
@@ -9456,7 +9622,7 @@ async function fetchEmbedModel(url, expectedSha256) {
   if (!res.ok) throw new Error(`HTTP ${res.status} from ${url}`);
   const body2 = await res.text();
   if (expectedSha256) {
-    const got = createHash2("sha256").update(body2).digest("hex");
+    const got = createHash3("sha256").update(body2).digest("hex");
     if (got !== expectedSha256) {
       throw new Error(`sha256 mismatch: expected ${expectedSha256}, got ${got}`);
     }
@@ -10060,8 +10226,8 @@ var init_repomap = __esm({
 
 // src/deadcode.ts
 function findDeadCode(scan2) {
-  const callers = buildCallerIndex(scan2);
-  const refs = computeSymbolRefs(scan2);
+  const callers = callerIndexFor(scan2);
+  const refs = symbolRefsFor(scan2);
   const out2 = [];
   const consider = (s) => s.exported && !REFERENCE_KINDS6.has(s.kind) && !isTestPath(s.file) && !ENTRYPOINT_RE.test(s.file);
   for (const f of scan2.files) {
@@ -10080,8 +10246,7 @@ var REFERENCE_KINDS6, ENTRYPOINT_RE;
 var init_deadcode = __esm({
   "src/deadcode.ts"() {
     "use strict";
-    init_callers();
-    init_symbols_json();
+    init_derived();
     init_tests_map();
     init_sort();
     REFERENCE_KINDS6 = /* @__PURE__ */ new Set(["reexport", "reexport-all", "default"]);
@@ -10089,49 +10254,6 @@ var init_deadcode = __esm({
   }
 });
 
-// src/complexity.ts
-import { join as join11 } from "path";
-function complexityOfSource(source) {
-  return 1 + (source.match(BRANCH_RE) ?? []).length;
-}
-function symbolComplexity(scan2, rel, top = 50) {
-  const out2 = [];
-  for (const f of scan2.files) {
-    if (f.kind !== "code") continue;
-    if (rel && f.rel !== rel) continue;
-    if (!f.symbols.length) continue;
-    const lines = readText(join11(scan2.root, f.rel)).split("\n");
-    for (const s of f.symbols) {
-      if (s.kind === "reexport" || s.kind === "reexport-all") continue;
-      const end = s.endLine ?? s.line;
-      const body2 = lines.slice(s.line - 1, end).join("\n");
-      const entry = { file: f.rel, name: s.name, line: s.line, complexity: complexityOfSource(body2) };
-      if (s.endLine !== void 0) entry.endLine = s.endLine;
-      out2.push(entry);
-    }
-  }
-  out2.sort((a, b) => b.complexity - a.complexity || byStr(a.file, b.file) || a.line - b.line);
-  return out2.slice(0, top);
-}
-function riskHotspots(scan2, churn, top = 20) {
-  const out2 = scan2.files.filter((f) => f.kind === "code").map((f) => {
-    const complexity = complexityOfSource(readText(join11(scan2.root, f.rel)));
-    const commits = churn.get(f.rel) ?? 0;
-    return { file: f.rel, complexity, commits, score: (commits + 1) * complexity };
-  });
-  out2.sort((a, b) => b.score - a.score || byStr(a.file, b.file));
-  return out2.slice(0, top);
-}
-var BRANCH_RE;
-var init_complexity = __esm({
-  "src/complexity.ts"() {
-    "use strict";
-    init_walk();
-    init_sort();
-    BRANCH_RE = /\b(if|elif|elsif|else\s+if|for|foreach|while|until|unless|case|when|match|catch|rescue|except)\b|&&|\|\||(?<![?:])\?(?![?.:])/g;
-  }
-});
-
 // src/viz.ts
 function renderMermaid(graph, opts = {}) {
   const maxEdges = opts.maxEdges ?? 80;
@@ -10172,10 +10294,17 @@ var init_viz = __esm({
 // src/mcp.ts
 var mcp_exports = {};
 __export(mcp_exports, {
+  getArtifacts: () => getArtifacts,
+  getScan: () => getScan,
+  memoizedEmbedModel: () => memoizedEmbedModel,
   memoizedEmbeddingIndex: () => memoizedEmbeddingIndex,
   runMcpServer: () => runMcpServer,
-  scanFingerprint: () => scanFingerprint
+  scanFingerprint: () => scanFingerprint,
+  toCacheMap: () => toCacheMap,
+  warmGrammarsForRepo: () => warmGrammarsForRepo
 });
+import { statSync as statSync4 } from "fs";
+import { join as join14 } from "path";
 import { createInterface } from "readline";
 function str(v) {
   return typeof v === "string" && v ? v : void 0;
@@ -10196,12 +10325,72 @@ async function memoizedEmbeddingIndex(key, build) {
   embeddingIndexCache = { key: cacheKey, index };
   return index;
 }
+function memoizedEmbedModel(modelDir) {
+  let stat;
+  try {
+    stat = statSync4(join14(modelDir, "model.json"));
+  } catch {
+    return void 0;
+  }
+  const key = `${modelDir}:${stat.mtimeMs}:${stat.size}`;
+  if (embedModelCache && embedModelCache.key === key) return embedModelCache.model;
+  const model = loadEmbedModel(modelDir);
+  if (model) embedModelCache = { key, model };
+  return model;
+}
+function sessionKey(repo, opts) {
+  return repo + "\0" + JSON.stringify({
+    scope: opts.scope,
+    include: opts.include,
+    exclude: opts.exclude,
+    gitignore: opts.gitignore,
+    ignoreDirs: opts.ignoreDirs,
+    maxBytes: opts.maxBytes,
+    maxFiles: opts.maxFiles,
+    maxCallsPerFile: opts.maxCallsPerFile,
+    out: opts.out,
+    fullHash: opts.fullHash
+  });
+}
+function toCacheMap(scan2) {
+  const m = /* @__PURE__ */ new Map();
+  for (const f of scan2.files) m.set(f.rel, { hash: f.hash, record: f, size: f.size, mtimeMs: scan2.mtimes.get(f.rel) });
+  return m;
+}
+function getScan(repo, opts = {}) {
+  const key = sessionKey(repo, opts);
+  if (sessionCache && sessionCache.key === key) {
+    const fresh = scanRepo(repo, { ...opts, cache: sessionCache.cacheMap });
+    if (fresh.contentUnchanged) {
+      if (fresh.cacheDirty) sessionCache.cacheMap = toCacheMap(fresh);
+      if (sessionCache.scan.commit !== fresh.commit) sessionCache.scan.commit = fresh.commit;
+      return sessionCache.scan;
+    }
+    sessionCache = { key, scan: fresh, cacheMap: toCacheMap(fresh) };
+    return fresh;
+  }
+  const scan2 = scanRepo(repo, opts);
+  sessionCache = { key, scan: scan2, cacheMap: toCacheMap(scan2) };
+  return scan2;
+}
+function getArtifacts(repo, opts = {}) {
+  const scan2 = getScan(repo, opts);
+  if (sessionCache && sessionCache.scan === scan2) {
+    return sessionCache.arts ??= buildArtifactsFromScan(scan2, opts);
+  }
+  return buildArtifactsFromScan(scan2, opts);
+}
+async function warmGrammarsForRepo(repo) {
+  const { files } = walk(repo, {});
+  await ensureGrammars(grammarKeysForExts(files.map((f) => f.ext)));
+}
 async function callTool(name2, args2) {
   const repo = str(args2.repo);
   if (!repo) throw new Error("`repo` is required (absolute path to the repository root)");
   const scanOpts = { scope: str(args2.scope), include: strArray(args2.include), exclude: strArray(args2.exclude) };
+  if (!SCANLESS_TOOLS.has(name2)) await warmGrammarsForRepo(repo);
   if (name2 === "scan_summary") {
-    const scan2 = scanRepo(repo, scanOpts);
+    const scan2 = getScan(repo, scanOpts);
     return JSON.stringify(
       { engineVersion: ENGINE_VERSION, commit: scan2.commit, fileCount: scan2.files.length, languages: scan2.languages, capped: scan2.capped },
       null,
@@ -10209,10 +10398,10 @@ async function callTool(name2, args2) {
     );
   }
   if (name2 === "graph") {
-    return renderGraphJson(buildIndexArtifacts(repo, scanOpts).graph);
+    return renderGraphJson(getArtifacts(repo, scanOpts).graph);
   }
   if (name2 === "symbols") {
-    const { symbols } = buildIndexArtifacts(repo, scanOpts);
+    const { symbols } = getArtifacts(repo, scanOpts);
     const lookup = str(args2.name);
     if (lookup) {
       return JSON.stringify({ name: lookup, defs: symbols.defs[lookup] ?? [], refs: symbols.refs[lookup] ?? [] }, null, 2);
@@ -10220,7 +10409,7 @@ async function callTool(name2, args2) {
     return JSON.stringify(symbols, null, 2);
   }
   if (name2 === "callers") {
-    const index = buildCallerIndex(scanRepo(repo, scanOpts));
+    const index = buildCallerIndex(getScan(repo, scanOpts));
     const lookup = str(args2.name);
     if (lookup) {
       const entry = index.get(lookup);
@@ -10243,12 +10432,12 @@ async function callTool(name2, args2) {
   if (name2 === "symbols_overview") {
     const file = str(args2.file);
     if (!file) throw new Error("`file` is required");
-    return JSON.stringify(symbolsOverview(scanRepo(repo, scanOpts), file), null, 2);
+    return JSON.stringify(symbolsOverview(getScan(repo, scanOpts), file), null, 2);
   }
   if (name2 === "find_symbol") {
     const namePath = str(args2.namePath);
     if (!namePath) throw new Error("`namePath` is required");
-    const matches = findSymbol(scanRepo(repo, scanOpts), namePath, {
+    const matches = findSymbol(getScan(repo, scanOpts), namePath, {
       substring: args2.substring === true,
       includeBody: args2.includeBody === true
     });
@@ -10257,15 +10446,17 @@ async function callTool(name2, args2) {
   if (name2 === "find_references") {
     const symName = str(args2.name);
     if (!symName) throw new Error("`name` is required");
-    return JSON.stringify(findReferences(scanRepo(repo, scanOpts), symName), null, 2);
+    return JSON.stringify(findReferences(getScan(repo, scanOpts), symName), null, 2);
   }
   if (name2 === "replace_symbol_body" || name2 === "insert_after_symbol" || name2 === "insert_before_symbol") {
     const namePath = str(args2.namePath);
     const body2 = typeof args2.body === "string" ? args2.body : void 0;
     if (!namePath || body2 === void 0) throw new Error("`namePath` and `body` are required");
-    const scan2 = scanRepo(repo, scanOpts);
+    const scan2 = getScan(repo, scanOpts);
     const fn = name2 === "replace_symbol_body" ? replaceSymbolBody : name2 === "insert_after_symbol" ? insertAfterSymbol : insertBeforeSymbol;
-    return JSON.stringify(fn(scan2, namePath, body2, str(args2.file)), null, 2);
+    const result = fn(scan2, namePath, body2, str(args2.file));
+    sessionCache = void 0;
+    return JSON.stringify(result, null, 2);
   }
   if (name2 === "write_memory") {
     const memName = str(args2.name);
@@ -10289,10 +10480,10 @@ async function callTool(name2, args2) {
     return JSON.stringify({ deleted: deleteMemory(repo, memName) }, null, 2);
   }
   if (name2 === "dead_code") {
-    return JSON.stringify(findDeadCode(scanRepo(repo, scanOpts)), null, 2);
+    return JSON.stringify(findDeadCode(getScan(repo, scanOpts)), null, 2);
   }
   if (name2 === "complexity") {
-    const scan2 = scanRepo(repo, scanOpts);
+    const scan2 = getScan(repo, scanOpts);
     if (args2.risk === true) {
       const { churn, ok } = gitChurn(repo);
       return JSON.stringify({ churnOk: ok, risks: riskHotspots(scan2, churn) }, null, 2);
@@ -10300,15 +10491,15 @@ async function callTool(name2, args2) {
     return JSON.stringify(symbolComplexity(scan2, str(args2.file)), null, 2);
   }
   if (name2 === "mermaid") {
-    const { graph } = buildIndexArtifacts(repo, scanOpts);
+    const { graph } = getArtifacts(repo, scanOpts);
     return renderMermaid(graph, { module: str(args2.module) });
   }
   if (name2 === "repo_map") {
-    const { scan: scan2, graph } = buildIndexArtifacts(repo, scanOpts);
+    const { scan: scan2, graph } = getArtifacts(repo, scanOpts);
     return renderRepoMap(scan2, graph, { budgetTokens: typeof args2.budgetTokens === "number" ? args2.budgetTokens : void 0 });
   }
   if (name2 === "hotspots") {
-    const scan2 = scanRepo(repo, scanOpts);
+    const scan2 = getScan(repo, scanOpts);
     const { churn, ok } = gitChurn(repo, { since: str(args2.since) });
     return JSON.stringify({ churnOk: ok, hotspots: rankHotspots(scan2, churn) }, null, 2);
   }
@@ -10329,7 +10520,7 @@ async function callTool(name2, args2) {
   if (name2 === "search") {
     const query = str(args2.query);
     if (!query) throw new Error("`query` is required");
-    const scan2 = scanRepo(repo, scanOpts);
+    const scan2 = getScan(repo, scanOpts);
     const limit = typeof args2.limit === "number" ? args2.limit : void 0;
     const fuzzy = typeof args2.fuzzy === "boolean" ? args2.fuzzy : void 0;
     if (args2.semantic === true) {
@@ -10350,7 +10541,7 @@ async function callTool(name2, args2) {
         }
       }
       const modelDir = resolveEmbedModelDir(repo);
-      const model = modelDir ? loadEmbedModel(modelDir) : void 0;
+      const model = modelDir ? memoizedEmbedModel(modelDir) : void 0;
       if (model) {
         const index = await memoizedEmbeddingIndex(
           { mode: "static", identity: `${modelDir}#${model.modelId}`, scan: scan2 },
@@ -10370,7 +10561,7 @@ async function callTool(name2, args2) {
   }
   if (name2 === "embed_status") {
     const modelDir = resolveEmbedModelDir(repo);
-    const model = modelDir ? loadEmbedModel(modelDir) : void 0;
+    const model = modelDir ? memoizedEmbedModel(modelDir) : void 0;
     const endpoint = resolveEmbedEndpoint();
     const mode = endpoint ? "endpoint" : model ? "static" : "none";
     const status = {
@@ -10384,13 +10575,16 @@ async function callTool(name2, args2) {
   }
   if (name2 === "check_rules") {
     const rules = parseRules(args2.rules);
-    const { graph } = buildIndexArtifacts(repo, scanOpts);
+    const { graph } = getArtifacts(repo, scanOpts);
     return JSON.stringify(checkRules(graph, rules), null, 2);
   }
   throw new Error(`unknown tool: ${name2}`);
 }
-async function runMcpServer() {
-  await ensureGrammars(allGrammarKeys());
+async function runMcpServer(opts = {}) {
+  const serverInfo = {
+    name: opts.serverInfo?.name ?? "codeindex",
+    version: opts.serverInfo?.version ?? ENGINE_VERSION
+  };
   const send = (msg) => {
     process.stdout.write(JSON.stringify({ jsonrpc: "2.0", ...msg }) + "\n");
   };
@@ -10417,7 +10611,7 @@ async function runMcpServer() {
           result: {
             protocolVersion: "2024-11-05",
             capabilities: { tools: {} },
-            serverInfo: { name: "codeindex", version: ENGINE_VERSION }
+            serverInfo
           }
         });
       } else if (req.method === "ping") {
@@ -10445,7 +10639,7 @@ async function runMcpServer() {
     }
   }
 }
-var repoProp, scopeProps, TOOLS, embeddingIndexCache;
+var repoProp, scopeProps, TOOLS, embeddingIndexCache, embedModelCache, sessionCache, SCANLESS_TOOLS;
 var init_mcp = __esm({
   "src/mcp.ts"() {
     "use strict";
@@ -10454,6 +10648,7 @@ var init_mcp = __esm({
     init_pipeline();
     init_graph_json();
     init_scan();
+    init_walk();
     init_callers();
     init_workspaces();
     init_git();
@@ -10724,6 +10919,17 @@ var init_mcp = __esm({
         }
       }
     ];
+    SCANLESS_TOOLS = /* @__PURE__ */ new Set([
+      "workspaces",
+      "churn",
+      "coupling",
+      "grep",
+      "write_memory",
+      "read_memory",
+      "list_memories",
+      "delete_memory",
+      "embed_status"
+    ]);
   }
 });
 
@@ -10863,6 +11069,106 @@ init_code();
 init_markdown();
 init_loader();
 init_extract();
+init_loader();
+
+// src/ast/grammars-pull.ts
+init_types();
+import { createHash as createHash2 } from "crypto";
+import { mkdirSync, writeFileSync } from "fs";
+import { dirname as dirname2, resolve, sep as sep2 } from "path";
+import { gunzipSync } from "zlib";
+var DEFAULT_GRAMMARS_URL = `https://github.com/maxgfr/codeindex/releases/download/v${ENGINE_VERSION}/grammars-${ENGINE_VERSION}.tar.gz`;
+function resolveGrammarsPullTarget() {
+  const env = process.env.CODEINDEX_GRAMMARS_URL;
+  if (env && env.trim()) return { url: env.trim() };
+  return { url: DEFAULT_GRAMMARS_URL, sha256Url: `${DEFAULT_GRAMMARS_URL}.sha256` };
+}
+async function fetchGrammarsTarball(url, expectedSha256) {
+  const res = await fetch(url);
+  if (!res.ok) throw new Error(`HTTP ${res.status} from ${url}`);
+  const buf = Buffer.from(await res.arrayBuffer());
+  if (expectedSha256) {
+    const got = createHash2("sha256").update(buf).digest("hex");
+    if (got !== expectedSha256) {
+      throw new Error(`sha256 mismatch: expected ${expectedSha256}, got ${got}`);
+    }
+  }
+  return buf;
+}
+function asBuffer(u) {
+  return Buffer.isBuffer(u) ? u : Buffer.from(u.buffer, u.byteOffset, u.byteLength);
+}
+async function fetchExpectedSha256(url) {
+  const res = await fetch(url);
+  if (!res.ok) throw new Error(`HTTP ${res.status} from ${url}`);
+  const text = await res.text();
+  const hex = (text.trim().split(/\s+/)[0] ?? "").toLowerCase();
+  if (!/^[0-9a-f]{64}$/.test(hex)) throw new Error(`invalid sha256 sidecar at ${url}`);
+  return hex;
+}
+function cstr(block, start2, len) {
+  const slice = block.subarray(start2, start2 + len);
+  const nul = slice.indexOf(0);
+  return slice.toString("utf8", 0, nul === -1 ? slice.length : nul);
+}
+function* readTar(buf) {
+  let off = 0;
+  while (off + 512 <= buf.length) {
+    const block = buf.subarray(off, off + 512);
+    let allZero = true;
+    for (let i2 = 0; i2 < 512; i2++) {
+      if (block[i2] !== 0) {
+        allZero = false;
+        break;
+      }
+    }
+    if (allZero) break;
+    const name2 = cstr(block, 0, 100);
+    const prefix = cstr(block, 345, 155);
+    const sizeStr = cstr(block, 124, 12).trim();
+    const size = sizeStr ? parseInt(sizeStr, 8) : 0;
+    const type = String.fromCharCode(block[156] ?? 0);
+    off += 512;
+    const data = buf.subarray(off, off + size);
+    off += Math.ceil(size / 512) * 512;
+    yield { name: prefix ? `${prefix}/${name2}` : name2, type, data };
+  }
+}
+function safeRelPath(name2) {
+  if (!name2 || name2.includes("\0")) return null;
+  if (name2.startsWith("/") || name2.startsWith("\\") || /^[A-Za-z]:/.test(name2)) return null;
+  const out2 = [];
+  for (const part of name2.split(/[/\\]/)) {
+    if (part === "" || part === ".") continue;
+    if (part === "..") return null;
+    out2.push(part);
+  }
+  return out2.length ? out2.join("/") : null;
+}
+function extractTarInto(rawTar, destDir) {
+  const root = resolve(destDir);
+  const written = [];
+  for (const entry of readTar(asBuffer(rawTar))) {
+    if (entry.type !== "0" && entry.type !== "\0") continue;
+    const rel = safeRelPath(entry.name);
+    if (rel === null) throw new Error(`refusing unsafe tar entry: ${entry.name}`);
+    const dest = resolve(destDir, rel);
+    if (dest !== root && !dest.startsWith(root + sep2)) {
+      throw new Error(`tar entry escapes destination: ${entry.name}`);
+    }
+    mkdirSync(dirname2(dest), { recursive: true });
+    writeFileSync(dest, entry.data);
+    written.push(rel);
+  }
+  return written;
+}
+function extractGrammarsTarball(bytes, destDir) {
+  const b = asBuffer(bytes);
+  const raw = b.length >= 2 && b[0] === 31 && b[1] === 139 ? gunzipSync(b) : b;
+  return extractTarInto(raw, destDir);
+}
+
+// src/engine.ts
 init_resolve();
 init_modules();
 init_graph();
@@ -10883,7 +11189,7 @@ init_graph_json();
 init_types();
 init_walk();
 init_sort();
-import { join as join9 } from "path";
+import { join as join12 } from "path";
 var utf8 = new TextEncoder();
 function pushVarint(out2, n) {
   if (n < 0) throw new Error(`pushVarint: negative input ${n} is not a valid unsigned varint`);
@@ -11053,7 +11359,7 @@ function renderScip(scan2, opts = {}) {
   };
   const documents = [];
   for (const f of docs) {
-    const text = readText(join9(scan2.root, f.rel));
+    const text = readText(join12(scan2.root, f.rel));
     const lines = text.split("\n").map((l) => l.endsWith("\r") ? l.slice(0, -1) : l);
     const locate = (lineNo, name2) => {
       const line = lines[lineNo - 1];
@@ -11144,12 +11450,14 @@ init_util();
 init_types();
 init_types();
 init_loader();
+import { existsSync as existsSync4, mkdirSync as mkdirSync3, mkdtempSync, readFileSync as readFileSync6, renameSync, rmSync as rmSync2, writeFileSync as writeFileSync4 } from "fs";
+import { dirname as dirname4, join as join15, resolve as resolve2 } from "path";
 init_pipeline();
+init_hash();
 init_graph_json();
 init_symbols_json();
-import { existsSync as existsSync4, mkdirSync as mkdirSync2, readFileSync as readFileSync6, writeFileSync as writeFileSync3 } from "fs";
-import { join as join12, resolve } from "path";
 init_scan();
+init_walk();
 init_callers();
 init_workspaces();
 init_git();
@@ -11196,6 +11504,14 @@ Commands:
                                the source with CODEINDEX_EMBED_URL
                 embed serve    Print (or --run) the docker command that starts the
                                containerized embedding server (rich tier)
+  grammars    Tree-sitter wasm grammars (optional AST tier; regex without them).
+              Precedence: bundle-adjacent > CODEINDEX_GRAMMARS_DIR > shared cache:
+                grammars status  Active tier (adjacent/env/cache/none), resolved
+                                 dir, pinned ENGINE_VERSION, pull-needed (JSON)
+                grammars pull    Fetch the per-release grammars-<version>.tar.gz
+                                 asset into the shared cache (sha256-verified,
+                                 atomic). Override the source with
+                                 CODEINDEX_GRAMMARS_URL
   rules       Architecture rules (forbidden edges, cycles, orphans) validated
               against the link-graph: --config <codeindex.rules.json>; exits 1
               on any error-severity violation (a CI gate)
@@ -11216,8 +11532,11 @@ Flags:
   --exclude <glob>    Exclude matching paths (repeatable)
   --scope <dir>       Restrict to one directory (sugar for --include '<dir>/**')
   --no-gitignore      Do not honor .gitignore files (default: honored)
+  --ignore-dir <name> Directory names to skip (repeatable) \u2014 REPLACES the
+                      default ignored-directory set, never merges with it
   --max-files <n>     Cap walked files (default 20000)
   --max-bytes <n>     Skip files above this size (default 1 MiB)
+  --max-calls <n>     Per-file call-site cap for extraction (default 512)
   --no-ast            Skip tree-sitter grammars even when present (regex tier)
   --config <file>     Rules config for \`rules\` (JSON: [{name, from, to, \u2026}])
   --limit <n>         Max results for \`search\` (default 20)
@@ -11232,7 +11551,7 @@ Flags:
                       each site corroborated|unique-name
 `;
 function parseFlags(args2) {
-  const flags2 = { repo: process.cwd(), include: [], exclude: [], gitignore: true, noAst: false, fuzzy: true, semantic: false };
+  const flags2 = { repo: process.cwd(), include: [], exclude: [], gitignore: true, ignoreDirs: [], noAst: false, fuzzy: true, semantic: false };
   for (let i2 = 0; i2 < args2.length; i2++) {
     const a = args2[i2];
     const next = () => {
@@ -11246,23 +11565,25 @@ function parseFlags(args2) {
       if (!Number.isFinite(n) || n <= 0) throw new Error(`${a} expects a positive number, got "${raw}"`);
       return n;
     };
-    if (a === "--repo") flags2.repo = resolve(next());
+    if (a === "--repo") flags2.repo = resolve2(next());
     else if (a === "--out") {
       const v = next();
-      flags2.out = v === "-" ? "-" : resolve(v);
+      flags2.out = v === "-" ? "-" : resolve2(v);
     } else if (a === "--project-root") flags2.projectRoot = next();
     else if (a === "--include") flags2.include.push(next());
     else if (a === "--exclude") flags2.exclude.push(next());
     else if (a === "--scope") flags2.scope = next();
     else if (a === "--no-gitignore") flags2.gitignore = false;
+    else if (a === "--ignore-dir") flags2.ignoreDirs.push(next());
     else if (a === "--max-files") flags2.maxFiles = num();
     else if (a === "--max-bytes") flags2.maxBytes = num();
+    else if (a === "--max-calls") flags2.maxCalls = num();
     else if (a === "--ignore-case") flags2.ignoreCase = true;
     else if (a === "--max-hits") flags2.maxHits = num();
     else if (a === "--budget-tokens") flags2.budgetTokens = num();
     else if (a === "--no-ast") flags2.noAst = true;
     else if (a === "--since") flags2.since = next();
-    else if (a === "--config") flags2.config = resolve(next());
+    else if (a === "--config") flags2.config = resolve2(next());
     else if (a === "--limit") flags2.limit = num();
     else if (a === "--no-fuzzy") flags2.fuzzy = false;
     else if (a === "--semantic") flags2.semantic = true;
@@ -11274,19 +11595,26 @@ function parseFlags(args2) {
   return flags2;
 }
 function emit(content, out2) {
-  if (out2) writeFileSync3(out2, content);
+  if (out2) writeFileSync4(out2, content);
   else process.stdout.write(content);
 }
-function scanOptions(flags2) {
+function scanOptions(flags2, precomputedWalk) {
   return {
     include: flags2.include.length ? flags2.include : void 0,
     exclude: flags2.exclude.length ? flags2.exclude : void 0,
     scope: flags2.scope,
     gitignore: flags2.gitignore,
+    ignoreDirs: flags2.ignoreDirs.length ? flags2.ignoreDirs : void 0,
     maxFiles: flags2.maxFiles,
-    maxBytes: flags2.maxBytes
+    maxBytes: flags2.maxBytes,
+    maxCallsPerFile: flags2.maxCalls,
+    // The walk performed once in runCli to warm the present-language grammars,
+    // reused here so scanRepo does not traverse the tree a second time. Absent
+    // for --no-ast / scan-less commands: scanRepo walks itself, unchanged.
+    precomputedWalk
   };
 }
+var SCANLESS_COMMANDS = /* @__PURE__ */ new Set(["grep", "churn", "coupling", "workspaces", "grammars"]);
 async function runCli(argv) {
   const [cmd, ...rest] = argv;
   if (!cmd || cmd === "help" || cmd === "--help" || cmd === "-h") {
@@ -11304,46 +11632,102 @@ async function runCli(argv) {
   }
   const flags2 = parseFlags(rest);
   if (!existsSync4(flags2.repo)) throw new Error(`--repo path does not exist: ${flags2.repo}`);
-  if (!flags2.noAst) await ensureGrammars(allGrammarKeys());
+  const scans = !SCANLESS_COMMANDS.has(cmd) && !(cmd === "embed" && flags2.positional !== "build");
+  let precomputedWalk;
+  if (scans && !flags2.noAst) {
+    precomputedWalk = walk(flags2.repo, {
+      maxFileBytes: flags2.maxBytes,
+      maxFiles: flags2.maxFiles,
+      gitignore: flags2.gitignore,
+      ignoreDirs: flags2.ignoreDirs.length ? flags2.ignoreDirs : void 0
+    });
+    await ensureGrammars(grammarKeysForExts(precomputedWalk.files.map((f) => f.ext)));
+  }
   if (cmd === "index") {
     if (!flags2.out) throw new Error("index needs --out <dir>");
     const outDir = flags2.out;
-    mkdirSync2(outDir, { recursive: true });
-    const cachePath = join12(outDir, "cache.json");
+    mkdirSync3(outDir, { recursive: true });
+    const cachePath = join15(outDir, "cache.json");
     let cache;
+    let meta = {};
     try {
       const parsed = JSON.parse(readFileSync6(cachePath, "utf8"));
       if (parsed.schemaVersion === SCHEMA_VERSION && parsed.extractorVersion === EXTRACTOR_VERSION) {
         cache = new Map(Object.entries(parsed.files));
+        meta = {
+          engineVersion: parsed.engineVersion,
+          commit: parsed.commit,
+          graphSha1: parsed.graphSha1,
+          symbolsSha1: parsed.symbolsSha1,
+          embed: parsed.embed
+        };
       }
     } catch {
     }
-    const { scan: scan2, graph, symbols } = buildIndexArtifacts(flags2.repo, { ...scanOptions(flags2), cache, out: outDir });
-    writeFileSync3(join12(outDir, "graph.json"), renderGraphJson(graph));
-    writeFileSync3(join12(outDir, "symbols.json"), renderSymbolsJson(symbols));
-    const files = {};
-    for (const f of scan2.files) {
-      const entry = { hash: f.hash, record: f, size: f.size };
-      const mtime = scan2.mtimes.get(f.rel);
-      if (mtime !== void 0) entry.mtimeMs = mtime;
-      files[f.rel] = entry;
-    }
-    writeFileSync3(
-      cachePath,
-      JSON.stringify({ schemaVersion: SCHEMA_VERSION, extractorVersion: EXTRACTOR_VERSION, files }) + "\n"
-    );
-    let embedNote = "";
+    const scan2 = scanRepo(flags2.repo, { ...scanOptions(flags2, precomputedWalk), cache, out: outDir });
     const modelDir = resolveEmbedModelDir(flags2.repo);
     const model = modelDir ? loadEmbedModel(modelDir) : void 0;
-    if (model) {
-      const index = buildEmbeddingIndex(scan2, model);
-      writeFileSync3(join12(outDir, "embeddings.bin"), serializeEmbeddings(index));
-      embedNote = ` + embeddings.bin (${index.records.length} records, model ${model.modelId})`;
-    }
-    process.stderr.write(`codeindex: ${scan2.files.length} files \u2192 ${outDir}/graph.json + symbols.json${embedNote}${scan2.capped ? " (capped)" : ""}
+    const graphPath = join15(outDir, "graph.json");
+    const symbolsPath = join15(outDir, "symbols.json");
+    const embedPath = join15(outDir, "embeddings.bin");
+    const artifactSha = (path) => {
+      try {
+        return sha1(readFileSync6(path));
+      } catch {
+        return void 0;
+      }
+    };
+    const writeCache = (out2) => {
+      const files = {};
+      for (const f of scan2.files) {
+        const entry = { hash: f.hash, record: f, size: f.size };
+        const mtime = scan2.mtimes.get(f.rel);
+        if (mtime !== void 0) entry.mtimeMs = mtime;
+        files[f.rel] = entry;
+      }
+      writeFileSync4(
+        cachePath,
+        JSON.stringify({
+          schemaVersion: SCHEMA_VERSION,
+          extractorVersion: EXTRACTOR_VERSION,
+          engineVersion: ENGINE_VERSION,
+          commit: scan2.commit,
+          graphSha1: out2.graphSha1,
+          symbolsSha1: out2.symbolsSha1,
+          embed: out2.embed,
+          files
+        }) + "\n"
+      );
+    };
+    const embedUnchanged = !model || meta.embed !== void 0 && meta.embed.embedVersion === EMBED_VERSION && meta.embed.modelId === model.modelId && meta.embed.sha1 !== void 0 && artifactSha(embedPath) === meta.embed.sha1;
+    const fastpath = scan2.contentUnchanged && meta.engineVersion === ENGINE_VERSION && meta.commit === scan2.commit && meta.graphSha1 !== void 0 && artifactSha(graphPath) === meta.graphSha1 && meta.symbolsSha1 !== void 0 && artifactSha(symbolsPath) === meta.symbolsSha1 && embedUnchanged;
+    if (fastpath) {
+      if (scan2.cacheDirty) writeCache(meta);
+      process.stderr.write(
+        `codeindex: ${scan2.files.length} files \u2192 ${outDir}/graph.json + symbols.json${scan2.capped ? " (capped)" : ""} (unchanged \u2014 artifacts reused)
+`
+      );
+    } else {
+      const { graph, symbols } = buildArtifactsFromScan(scan2);
+      const graphJson = renderGraphJson(graph);
+      const symbolsJson = renderSymbolsJson(symbols);
+      writeFileSync4(graphPath, graphJson);
+      writeFileSync4(symbolsPath, symbolsJson);
+      let embedNote = "";
+      let embedMeta;
+      if (model) {
+        const index = buildEmbeddingIndex(scan2, model);
+        const bytes = serializeEmbeddings(index);
+        writeFileSync4(embedPath, bytes);
+        embedMeta = { embedVersion: EMBED_VERSION, modelId: model.modelId, sha1: sha1(bytes) };
+        embedNote = ` + embeddings.bin (${index.records.length} records, model ${model.modelId})`;
+      }
+      writeCache({ graphSha1: sha1(graphJson), symbolsSha1: sha1(symbolsJson), embed: embedMeta });
+      process.stderr.write(`codeindex: ${scan2.files.length} files \u2192 ${outDir}/graph.json + symbols.json${embedNote}${scan2.capped ? " (capped)" : ""}
 `);
+    }
   } else if (cmd === "scan") {
-    const { scan: scan2 } = buildIndexArtifacts(flags2.repo, scanOptions(flags2));
+    const { scan: scan2 } = buildIndexArtifacts(flags2.repo, scanOptions(flags2, precomputedWalk));
     const summary = {
       engineVersion: ENGINE_VERSION,
       commit: scan2.commit,
@@ -11353,30 +11737,30 @@ async function runCli(argv) {
     };
     emit(JSON.stringify(summary, null, 2) + "\n", flags2.out);
   } else if (cmd === "graph") {
-    const { graph } = buildIndexArtifacts(flags2.repo, scanOptions(flags2));
+    const { graph } = buildIndexArtifacts(flags2.repo, scanOptions(flags2, precomputedWalk));
     emit(renderGraphJson(graph), flags2.out);
   } else if (cmd === "symbols") {
-    const { symbols } = buildIndexArtifacts(flags2.repo, scanOptions(flags2));
+    const { symbols } = buildIndexArtifacts(flags2.repo, scanOptions(flags2, precomputedWalk));
     emit(renderSymbolsJson(symbols), flags2.out);
   } else if (cmd === "scip") {
-    const scan2 = scanRepo(flags2.repo, scanOptions(flags2));
+    const scan2 = scanRepo(flags2.repo, scanOptions(flags2, precomputedWalk));
     const bytes = renderScip(scan2, { projectRoot: flags2.projectRoot });
-    const out2 = flags2.out ?? resolve("index.scip");
+    const out2 = flags2.out ?? resolve2("index.scip");
     if (out2 === "-") process.stdout.write(Buffer.from(bytes));
     else {
-      writeFileSync3(out2, bytes);
+      writeFileSync4(out2, bytes);
       process.stderr.write(`codeindex: SCIP index \u2192 ${out2} (${bytes.length} bytes)
 `);
     }
   } else if (cmd === "callers") {
-    const scan2 = scanRepo(flags2.repo, scanOptions(flags2));
+    const scan2 = scanRepo(flags2.repo, scanOptions(flags2, precomputedWalk));
     const index = buildCallerIndex(scan2, void 0, { recall: flags2.recall });
     const obj = {};
     for (const [name2, entry] of index) obj[name2] = entry;
     emit(JSON.stringify(obj, null, 2) + "\n", flags2.out);
   } else if (cmd === "search") {
     if (!flags2.positional) throw new Error('search needs a query: cli.mjs search "<query>" --repo <dir>');
-    const scan2 = scanRepo(flags2.repo, scanOptions(flags2));
+    const scan2 = scanRepo(flags2.repo, scanOptions(flags2, precomputedWalk));
     if (flags2.semantic) {
       const endpoint = resolveEmbedEndpoint();
       const lexical = () => {
@@ -11466,17 +11850,17 @@ async function runCli(argv) {
         return;
       }
       const model = loadEmbedModel(modelDir);
-      mkdirSync2(flags2.out, { recursive: true });
-      const scan2 = scanRepo(flags2.repo, scanOptions(flags2));
+      mkdirSync3(flags2.out, { recursive: true });
+      const scan2 = scanRepo(flags2.repo, scanOptions(flags2, precomputedWalk));
       const index = buildEmbeddingIndex(scan2, model);
-      writeFileSync3(join12(flags2.out, "embeddings.bin"), serializeEmbeddings(index));
+      writeFileSync4(join15(flags2.out, "embeddings.bin"), serializeEmbeddings(index));
       process.stderr.write(`codeindex: ${index.records.length} embedding records \u2192 ${flags2.out}/embeddings.bin (model ${model.modelId})
 `);
     } else if (sub === "pull") {
       const { url, sha256 } = resolveEmbedPullUrl();
-      const destDir = process.env.CODEINDEX_EMBED_DIR ?? join12(flags2.repo, ".codeindex", "models");
-      mkdirSync2(destDir, { recursive: true });
-      process.stderr.write(`codeindex: fetching model from ${url} \u2192 ${join12(destDir, "model.json")}
+      const destDir = process.env.CODEINDEX_EMBED_DIR ?? join15(flags2.repo, ".codeindex", "models");
+      mkdirSync3(destDir, { recursive: true });
+      process.stderr.write(`codeindex: fetching model from ${url} \u2192 ${join15(destDir, "model.json")}
 `);
       let body2;
       try {
@@ -11488,22 +11872,111 @@ async function runCli(argv) {
         return;
       }
       try {
-        JSON.parse(body2);
-      } catch {
-        process.stderr.write("codeindex: pull failed \u2014 response is not a valid model.json (expected JSON)\n");
+        parseEmbedModel(JSON.parse(body2), url);
+      } catch (e) {
+        process.stderr.write(
+          `codeindex: pull failed \u2014 response is not a valid model.json (${e instanceof Error ? e.message : String(e)}) (nothing written)
+`
+        );
         process.exitCode = 1;
         return;
       }
-      writeFileSync3(join12(destDir, "model.json"), body2);
-      process.stderr.write(`codeindex: model written to ${join12(destDir, "model.json")}
+      writeFileSync4(join15(destDir, "model.json"), body2);
+      process.stderr.write(`codeindex: model written to ${join15(destDir, "model.json")}
 `);
     } else {
       throw new Error("embed needs a subcommand: status | build | pull | serve");
     }
+  } else if (cmd === "grammars") {
+    const sub = flags2.positional;
+    const cacheDir = sharedGrammarsCacheDir();
+    if (sub === "status") {
+      const info2 = resolveGrammarsTier();
+      const runtimePresent = info2.dir ? existsSync4(join15(info2.dir, "web-tree-sitter.wasm")) : false;
+      const target = resolveGrammarsPullTarget();
+      const status = {
+        engineVersion: ENGINE_VERSION,
+        tier: info2.tier,
+        dir: info2.dir ?? null,
+        cacheDir,
+        runtimePresent,
+        pullNeeded: !runtimePresent,
+        url: target.url
+      };
+      emit(JSON.stringify(status, null, 2) + "\n", flags2.out);
+    } else if (sub === "pull") {
+      const target = resolveGrammarsPullTarget();
+      let expected;
+      if (target.sha256Url) {
+        try {
+          expected = await fetchExpectedSha256(target.sha256Url);
+        } catch (e) {
+          process.stderr.write(
+            `codeindex: could not fetch checksum (${e instanceof Error ? e.message : String(e)}) \u2014 proceeding unverified
+`
+          );
+        }
+      }
+      const runtime = join15(cacheDir, "web-tree-sitter.wasm");
+      const markerPath = join15(dirname4(cacheDir), `${ENGINE_VERSION}.sha256`);
+      if (existsSync4(runtime) && expected && existsSync4(markerPath)) {
+        let marker = "";
+        try {
+          marker = readFileSync6(markerPath, "utf8").trim();
+        } catch {
+        }
+        if (marker === expected) {
+          process.stderr.write(`codeindex: grammars already present at ${cacheDir} (up to date)
+`);
+          return;
+        }
+      }
+      process.stderr.write(`codeindex: fetching grammars from ${target.url} \u2192 ${cacheDir}
+`);
+      let bytes;
+      try {
+        bytes = await fetchGrammarsTarball(target.url, expected);
+      } catch (e) {
+        process.stderr.write(`codeindex: pull failed \u2014 ${e instanceof Error ? e.message : String(e)} (nothing written)
+`);
+        process.exitCode = 1;
+        return;
+      }
+      let tmp;
+      try {
+        mkdirSync3(dirname4(cacheDir), { recursive: true });
+        tmp = mkdtempSync(join15(dirname4(cacheDir), ".grammars-tmp-"));
+        extractGrammarsTarball(bytes, tmp);
+        if (!existsSync4(join15(tmp, "web-tree-sitter.wasm"))) {
+          throw new Error("archive is missing web-tree-sitter.wasm");
+        }
+        if (existsSync4(cacheDir)) rmSync2(cacheDir, { recursive: true, force: true });
+        renameSync(tmp, cacheDir);
+        tmp = void 0;
+        if (expected) writeFileSync4(markerPath, expected + "\n");
+      } catch (e) {
+        if (tmp) {
+          try {
+            rmSync2(tmp, { recursive: true, force: true });
+          } catch {
+          }
+        }
+        process.stderr.write(
+          `codeindex: pull failed \u2014 ${e instanceof Error ? e.message : String(e)} (nothing written)
+`
+        );
+        process.exitCode = 1;
+        return;
+      }
+      process.stderr.write(`codeindex: grammars extracted \u2192 ${cacheDir}
+`);
+    } else {
+      throw new Error("grammars needs a subcommand: status | pull");
+    }
   } else if (cmd === "rules") {
     if (!flags2.config) throw new Error("rules needs --config <codeindex.rules.json>");
     const rules = parseRules(JSON.parse(readFileSync6(flags2.config, "utf8")));
-    const { graph } = buildIndexArtifacts(flags2.repo, scanOptions(flags2));
+    const { graph } = buildIndexArtifacts(flags2.repo, scanOptions(flags2, precomputedWalk));
     const violations = checkRules(graph, rules);
     const errors = violations.filter((v) => v.severity === "error").length;
     emit(JSON.stringify({ errors, warnings: violations.length - errors, violations }, null, 2) + "\n", flags2.out);
@@ -11524,26 +11997,26 @@ async function runCli(argv) {
     for (const k of [...churn.keys()].sort()) sorted[k] = churn.get(k);
     emit(JSON.stringify({ ok, churn: sorted }, null, 2) + "\n", flags2.out);
   } else if (cmd === "repomap") {
-    const { scan: scan2, graph } = buildIndexArtifacts(flags2.repo, scanOptions(flags2));
+    const { scan: scan2, graph } = buildIndexArtifacts(flags2.repo, scanOptions(flags2, precomputedWalk));
     emit(renderRepoMap(scan2, graph, { budgetTokens: flags2.budgetTokens }), flags2.out);
   } else if (cmd === "hotspots") {
-    const scan2 = scanRepo(flags2.repo, scanOptions(flags2));
+    const scan2 = scanRepo(flags2.repo, scanOptions(flags2, precomputedWalk));
     const { churn, ok } = gitChurn(flags2.repo, { since: flags2.since });
     emit(JSON.stringify({ churnOk: ok, hotspots: rankHotspots(scan2, churn) }, null, 2) + "\n", flags2.out);
   } else if (cmd === "coupling") {
     const { ok, couplings } = changeCoupling(flags2.repo, { since: flags2.since });
     emit(JSON.stringify({ ok, couplings }, null, 2) + "\n", flags2.out);
   } else if (cmd === "deadcode") {
-    emit(JSON.stringify(findDeadCode(scanRepo(flags2.repo, scanOptions(flags2))), null, 2) + "\n", flags2.out);
+    emit(JSON.stringify(findDeadCode(scanRepo(flags2.repo, scanOptions(flags2, precomputedWalk))), null, 2) + "\n", flags2.out);
   } else if (cmd === "complexity") {
-    const scan2 = scanRepo(flags2.repo, scanOptions(flags2));
+    const scan2 = scanRepo(flags2.repo, scanOptions(flags2, precomputedWalk));
     emit(JSON.stringify(symbolComplexity(scan2, flags2.positional), null, 2) + "\n", flags2.out);
   } else if (cmd === "risk") {
-    const scan2 = scanRepo(flags2.repo, scanOptions(flags2));
+    const scan2 = scanRepo(flags2.repo, scanOptions(flags2, precomputedWalk));
     const { churn, ok } = gitChurn(flags2.repo, { since: flags2.since });
     emit(JSON.stringify({ churnOk: ok, risks: riskHotspots(scan2, churn) }, null, 2) + "\n", flags2.out);
   } else if (cmd === "mermaid") {
-    const { graph } = buildIndexArtifacts(flags2.repo, scanOptions(flags2));
+    const { graph } = buildIndexArtifacts(flags2.repo, scanOptions(flags2, precomputedWalk));
     emit(renderMermaid(graph, { module: flags2.positional }), flags2.out);
   } else if (cmd === "grep") {
     if (!flags2.positional) throw new Error("grep needs a pattern: cli.mjs grep <pattern> --repo <dir>");
@@ -11562,6 +12035,7 @@ ${HELP}`);
   }
 }
 export {
+  DEFAULT_GRAMMARS_URL,
   DEFAULT_MAX_FILES,
   EMBED_VERSION,
   ENGINE_VERSION,
@@ -11572,6 +12046,7 @@ export {
   applyCentrality,
   basicTokenize,
   betweennessOf,
+  buildArtifactsFromScan,
   buildCallerIndex,
   buildEmbeddingIndex,
   buildEndpointIndex,
@@ -11614,14 +12089,19 @@ export {
   extToLang,
   extractAst,
   extractCode,
+  extractGrammarsTarball,
   extractMarkdown,
   extractSymbols,
+  extractTarInto,
+  fetchExpectedSha256,
+  fetchGrammarsTarball,
   findDeadCode,
   findReferences,
   findSymbol,
   foldText,
   gitChurn,
   grammarKeyForExt,
+  grammarKeysForExts,
   grammarReady,
   grepRepo,
   hasEmbedModel,
@@ -11663,6 +12143,9 @@ export {
   resolveEmbedEndpoint,
   resolveEmbedModelDir,
   resolveEmbedPullUrl,
+  resolveGrammarsDir,
+  resolveGrammarsPullTarget,
+  resolveGrammarsTier,
   resolveImport,
   resolveUniqueSymbol,
   riskHotspots,
@@ -11676,6 +12159,7 @@ export {
   serializeEmbeddings,
   sh,
   sha1,
+  sharedGrammarsCacheDir,
   shortHash,
   slugify,
   subtokens,