fix(extraction): don't index linked git worktrees as embedded repos

CHANGELOG.md

@@ -9,6 +9,10 @@ and adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 ## [Unreleased]
 
+### Bug Fixes
+
+- **Linked git worktrees parked inside a repo are no longer indexed as separate projects.** When you keep `git worktree add` checkouts in a gitignored subdir (`.worktrees/`, `.claude/worktrees/`, …), the embedded-repo discovery (#514) treated each worktree as its own repo and re-indexed it — duplicating every symbol once per worktree (an N-worktree repo ballooned the index ~Nx and made `query`/`callers`/`impact` return N copies of each result). CodeGraph now detects a linked worktree (its `.git` is a file resolving into another repo's `worktrees/<id>` admin dir) and skips it, since it's the same repo as the main checkout. Submodules and genuine embedded repos are unaffected.
+
 
 ## [1.0.0] - 2026-06-12
 

__tests__/multi-repo-workspace.test.ts

@@ -64,6 +64,31 @@ describe('multi-repo workspaces (#514)', () => {
     expect(files).toContain('tools.ts'); // the parent's own tracked code still indexes
   });
 
+  it('excludes linked git worktrees nested under the repo, but keeps genuine embedded repos', () => {
+    // Main repo with a tracked file; .worktrees/ is gitignored (the common
+    // place tools put linked worktrees).
+    write(path.join(ws, 'src/app.ts'), 'export function app() { return 1; }\n');
+    write(path.join(ws, '.gitignore'), '/.worktrees/\n');
+    makeRepo(ws);
+
+    // A REAL linked worktree: its `.git` is a file pointing into
+    // ws/.git/worktrees/<id>, and it holds a full copy of the SAME repo's
+    // tracked files — re-indexing it duplicates every symbol N times.
+    git(ws, 'worktree', 'add', '-q', '-b', 'wt-branch', path.join(ws, '.worktrees/wt'));
+
+    // A genuine independent repo under the same ignored dir MUST still be
+    // discovered (#514 must not regress).
+    write(path.join(ws, '.worktrees/real-embed/src/lib.ts'), 'export function lib() {}\n');
+    makeRepo(path.join(ws, '.worktrees/real-embed'));
+
+    const files = scanDirectory(ws);
+    expect(files).toContain('src/app.ts'); // the main checkout indexes normally
+    // the linked worktree's copy is NOT re-indexed (no symbol duplication)
+    expect(files.some((f) => f.startsWith('.worktrees/wt/'))).toBe(false);
+    // a real embedded repo in the same ignored dir is still picked up
+    expect(files).toContain('.worktrees/real-embed/src/lib.ts');
+  });
+
   it('keeps respecting the parent .gitignore for the parent own (non-repo) dirs', () => {
     write(path.join(ws, 'scratch/junk.ts'), 'export function junk() { return 9; }\n');
     write(path.join(ws, 'src/app.ts'), 'export function app() { return 1; }\n');

src/extraction/index.ts

@@ -280,12 +280,37 @@ const EMBEDDED_REPO_SEARCH_DEPTH = 4;
 /** Max directories examined per search — a huge ignored data dir must never stall a scan/sync. */
 const EMBEDDED_REPO_SEARCH_ENTRIES = 2000;
 
+/**
+ * True if `dotGitPath` is a *linked git worktree* pointer — a `.git` FILE whose
+ * `gitdir:` resolves into another repo's `worktrees/<id>` admin dir. Such a
+ * worktree is a second checkout of an ALREADY-indexed repo (the main checkout),
+ * not an independent project, so the indexer must not re-walk it — doing so
+ * duplicates every symbol once per worktree. Workflows that park linked
+ * worktrees in a gitignored subdir (`.worktrees/`, `.claude/worktrees/`, …) are
+ * common, and the embedded-repo discovery (#514) would otherwise treat each as
+ * its own repo. A submodule pointer (`gitdir: …/modules/<id>`) and a genuine
+ * embedded repo (`.git` is a directory) both return false, so neither is
+ * affected.
+ */
+function isLinkedWorktree(dotGitPath: string): boolean {
+  try {
+    if (!fs.lstatSync(dotGitPath).isFile()) return false;
+    const m = /^gitdir:\s*(.+)$/m.exec(fs.readFileSync(dotGitPath, 'utf-8'));
+    if (!m) return false;
+    return /(^|[\\/])worktrees[\\/]/.test((m[1] ?? '').trim());
+  } catch {
+    return false;
+  }
+}
+
 /**
  * Find git repositories nested under `absDir` (inclusive), shallow bounded BFS.
  * Stops descending at each repo root found — contents belong to that repo's own
  * enumeration. Skips default-ignored dirs (`node_modules` can contain `.git`
- * from npm git-dependencies — that never makes it project code) and CodeGraph
- * data dirs. Depth- and entry-capped so a huge ignored tree can't stall the scan.
+ * from npm git-dependencies — that never makes it project code), CodeGraph
+ * data dirs, and linked git worktrees (same repo as the main checkout — see
+ * isLinkedWorktree). Depth- and entry-capped so a huge ignored tree can't stall
+ * the scan.
  */
 function findNestedGitRepos(absDir: string, relPrefix: string): string[] {
   const found: string[] = [];
@@ -300,8 +325,12 @@ function findNestedGitRepos(absDir: string, relPrefix: string): string[] {
       logDebug('Embedded-repo search entry cap hit — deeper repos (if any) not discovered', { under: relPrefix });
       break;
     }
-    if (fs.existsSync(path.join(abs, '.git'))) {
-      found.push(rel);
+    const dotGit = path.join(abs, '.git');
+    if (fs.existsSync(dotGit)) {
+      // A linked worktree is the same repo as the main checkout (already
+      // indexed) — record nothing so it isn't re-walked, but still stop
+      // descending: its contents belong to that worktree, not the parent.
+      if (!isLinkedWorktree(dotGit)) found.push(rel);
       continue; // its own git handles everything below
     }
     if (depth >= EMBEDDED_REPO_SEARCH_DEPTH) continue;