diff --git a/electron/main/index.ts b/electron/main/index.ts
index 1470b4c13..ad23187a8 100644
--- a/electron/main/index.ts
+++ b/electron/main/index.ts
@@ -197,12 +197,22 @@ async function initialize(): Promise<void> {
     mainWindow?.webContents.send('gateway:error', String(error));
   }
 
-  // Merge ClawX context snippets into the (now fully-seeded) bootstrap files
-  try {
-    ensureClawXContext();
-  } catch (error) {
+  // Merge ClawX context snippets into the workspace bootstrap files.
+  // The gateway seeds workspace files asynchronously after its HTTP server
+  // is ready, so ensureClawXContext will retry until the target files appear.
+  void ensureClawXContext().catch((error) => {
     logger.warn('Failed to merge ClawX context into workspace:', error);
-  }
+  });
+
+  // Re-apply ClawX context after every gateway restart because the gateway
+  // may re-seed workspace files with clean templates (losing ClawX markers).
+  gatewayManager.on('status', (status: { state: string }) => {
+    if (status.state === 'running') {
+      void ensureClawXContext().catch((error) => {
+        logger.warn('Failed to re-merge ClawX context after gateway reconnect:', error);
+      });
+    }
+  });
 }
 
 // Application lifecycle
diff --git a/electron/utils/openclaw-workspace.ts b/electron/utils/openclaw-workspace.ts
index d0973f409..02fe64db4 100644
--- a/electron/utils/openclaw-workspace.ts
+++ b/electron/utils/openclaw-workspace.ts
@@ -116,29 +116,26 @@ function resolveAllWorkspaceDirs(): string[] {
 }
 
 /**
- * Ensure ClawX context snippets are merged into the openclaw workspace
- * bootstrap files. Reads `*.clawx.md` templates from resources/context/
- * and injects them as marker-delimited sections into the corresponding
- * workspace `.md` files (e.g. AGENTS.clawx.md -> AGENTS.md).
- *
- * Iterates over every discovered agent workspace so all agents receive
- * the ClawX context regardless of which one is active.
+ * Synchronously merge ClawX context snippets into workspace bootstrap
+ * files that already exist on disk. Returns the number of target files
+ * that were skipped because they don't exist yet.
  */
-export function ensureClawXContext(): void {
+function mergeClawXContextOnce(): number {
   const contextDir = join(getResourcesDir(), 'context');
   if (!existsSync(contextDir)) {
     logger.debug('ClawX context directory not found, skipping context merge');
-    return;
+    return 0;
   }
 
   let files: string[];
   try {
     files = readdirSync(contextDir).filter((f) => f.endsWith('.clawx.md'));
   } catch {
-    return;
+    return 0;
   }
 
   const workspaceDirs = resolveAllWorkspaceDirs();
+  let skipped = 0;
 
   for (const workspaceDir of workspaceDirs) {
     if (!existsSync(workspaceDir)) {
@@ -151,6 +148,7 @@ export function ensureClawXContext(): void {
 
       if (!existsSync(targetPath)) {
         logger.debug(`Skipping ${targetName} in ${workspaceDir} (file does not exist yet, will be seeded by gateway)`);
+        skipped++;
         continue;
       }
 
@@ -164,4 +162,37 @@ export function ensureClawXContext(): void {
       }
     }
   }
+
+  return skipped;
+}
+
+const RETRY_INTERVAL_MS = 2000;
+const MAX_RETRIES = 15;
+
+/**
+ * Ensure ClawX context snippets are merged into the openclaw workspace
+ * bootstrap files. Reads `*.clawx.md` templates from resources/context/
+ * and injects them as marker-delimited sections into the corresponding
+ * workspace `.md` files (e.g. AGENTS.clawx.md -> AGENTS.md).
+ *
+ * The gateway seeds workspace files asynchronously after its HTTP server
+ * starts, so the target files may not exist yet when this is first called.
+ * When files are missing, retries with a delay until all targets are merged
+ * or the retry budget is exhausted.
+ */
+export async function ensureClawXContext(): Promise<void> {
+  let skipped = mergeClawXContextOnce();
+  if (skipped === 0) return;
+
+  for (let attempt = 1; attempt <= MAX_RETRIES; attempt++) {
+    await new Promise((r) => setTimeout(r, RETRY_INTERVAL_MS));
+    skipped = mergeClawXContextOnce();
+    if (skipped === 0) {
+      logger.info(`ClawX context merge completed after ${attempt} retry(ies)`);
+      return;
+    }
+    logger.debug(`ClawX context merge: ${skipped} file(s) still missing (retry ${attempt}/${MAX_RETRIES})`);
+  }
+
+  logger.warn(`ClawX context merge: ${skipped} file(s) still missing after ${MAX_RETRIES} retries`);
 }
diff --git a/resources/context/AGENTS.clawx.md b/resources/context/AGENTS.clawx.md
index a5f2231c5..e62b8684a 100644
--- a/resources/context/AGENTS.clawx.md
+++ b/resources/context/AGENTS.clawx.md
@@ -1,8 +1,3 @@
 ## ClawX Environment
 
-You are ClawX, a desktop AI assistant application based on OpenClaw.
-
-- **Python**: Always use `uv` to run Python commands. The `uv` binary is bundled and available on PATH. Examples: `uv run python script.py`, `uv pip install package`.
-- **Browser**: Full browser automation is available via the `browser` tool. The default "openclaw" profile uses an isolated browser instance. Use it for web scraping, form filling, testing, and any browser automation task. For simply opening a URL for the user to view, use `shell:openExternal` instead.
-- **Shell**: You have full shell access on the user's machine. Prefer using tools directly over asking the user to run commands manually.
-- Always confirm before running destructive operations.
+You are ClawX, a desktop AI assistant application based on OpenClaw. See TOOLS.md for ClawX-specific tool notes (uv, browser automation, etc.).
diff --git a/resources/context/TOOLS.clawx.md b/resources/context/TOOLS.clawx.md
index 359b054f2..bb48656ed 100644
--- a/resources/context/TOOLS.clawx.md
+++ b/resources/context/TOOLS.clawx.md
@@ -2,16 +2,12 @@
 
 ### uv (Python)
 
-- `uv` is the default Python environment manager. It is bundled with ClawX and on PATH.
-- Use `uv run python <script>` to execute Python scripts.
-- Use `uv pip install <package>` to install packages.
-- Do NOT use bare `python` or `pip` -- always go through `uv`.
+- `uv` is bundled with ClawX and on PATH. Do NOT use bare `python` or `pip`.
+- Run scripts: `uv run python <script>` | Install packages: `uv pip install <package>`
 
 ### Browser
 
-- The `browser` tool provides full browser automation via OpenClaw's browser control server.
-- Default profile is "openclaw" (isolated managed browser using system Chrome/Brave/Edge).
-- Use `action="start"` to launch the browser, then `action="snapshot"` to see the page, `action="act"` to interact.
-- Use `action="open"` with `targetUrl` to open new tabs.
-- Refs from snapshots (e.g. `e12`) are used in `act` actions to click/type on specific elements.
-- For simple "open a URL for the user to see", use `shell:openExternal` instead.
+- `browser` tool provides full automation (scraping, form filling, testing) via an isolated managed browser.
+- Flow: `action="start"` → `action="snapshot"` (see page + get element refs like `e12`) → `action="act"` (click/type using refs).
+- Open new tabs: `action="open"` with `targetUrl`.
+- To just open a URL for the user to view, use `shell:openExternal` instead.