diff --git a/src/cli/commands/serve.ts b/src/cli/commands/serve.ts
index 5d7c716..e4718ff 100644
--- a/src/cli/commands/serve.ts
+++ b/src/cli/commands/serve.ts
@@ -13,6 +13,7 @@ import {
   deriveCompletions,
   type ContentRegistry,
 } from "../content-registry.js";
+import { readContentIndex } from "../content-index.js";
 import type { CompletionsPayload } from "../completions/types.js";
 
 interface ContentIndex {
@@ -86,39 +87,7 @@ function getBestIP(): string {
  * 扫描目录内的 .md 等文件，构建内容注册表（路径索引 + 每文档内联内容）
  */
 export function buildRegistry(dir: string): ContentRegistry {
-  const index: Record<string, string> = {};
-
-  function scan(currentPath: string, relativePath: string) {
-    const entries = readdirSync(currentPath);
-
-    for (const entry of entries) {
-      if (entry.startsWith(".") || entry === "node_modules") continue;
-
-      const fullPath = join(currentPath, entry);
-      const relPath = relativePath ? join(relativePath, entry) : entry;
-      const normalizedRelPath = normalizePathKey(relPath);
-
-      const stats = statSync(fullPath);
-      if (stats.isDirectory()) {
-        scan(fullPath, relPath);
-      } else if (
-        entry.endsWith(".md") ||
-        entry.endsWith(".csv") ||
-        entry.endsWith(".yarn") ||
-        entry.endsWith(".svg")
-      ) {
-        try {
-          index[normalizedRelPath] = readFileSync(fullPath, "utf-8");
-        } catch (e) {
-          console.error(`读取文件失败：${fullPath}`, e);
-        }
-      }
-    }
-  }
-
-  scan(dir, "");
-
-  return buildRegistryFromIndex(index);
+  return buildRegistryFromIndex(readContentIndex(dir));
 }
 
 /**
diff --git a/src/cli/content-index.ts b/src/cli/content-index.ts
new file mode 100644
index 0000000..3c12a44
--- /dev/null
+++ b/src/cli/content-index.ts
@@ -0,0 +1,52 @@
+/**
+ * Filesystem walk that builds a raw `{ path: content }` index.
+ *
+ * Extracted from `buildRegistry` in `commands/serve.ts` so the validator can
+ * obtain the *raw* file contents (needed for accurate line numbers) while the
+ * server gets the scanned registry — both from the same walk.
+ */
+
+import { readdirSync, statSync, readFileSync } from "fs";
+import { join } from "path";
+import { normalizePathKey } from "./content-registry.js";
+
+export interface ContentIndex {
+  [path: string]: string;
+}
+
+/** Recursively read indexable files (`.md`, `.csv`, `.yarn`, `.svg`). */
+export function readContentIndex(dir: string): ContentIndex {
+  const index: ContentIndex = {};
+
+  function scan(currentPath: string, relativePath: string) {
+    const entries = readdirSync(currentPath);
+
+    for (const entry of entries) {
+      if (entry.startsWith(".") || entry === "node_modules") continue;
+
+      const fullPath = join(currentPath, entry);
+      const relPath = relativePath ? join(relativePath, entry) : entry;
+      const normalizedRelPath = normalizePathKey(relPath);
+
+      const stats = statSync(fullPath);
+      if (stats.isDirectory()) {
+        scan(fullPath, relPath);
+      } else if (
+        entry.endsWith(".md") ||
+        entry.endsWith(".csv") ||
+        entry.endsWith(".yarn") ||
+        entry.endsWith(".svg")
+      ) {
+        try {
+          index[normalizedRelPath] = readFileSync(fullPath, "utf-8");
+        } catch (e) {
+          console.error(`读取文件失败：${fullPath}`, e);
+        }
+      }
+    }
+  }
+
+  scan(dir, "");
+
+  return index;
+}