[eric] auto-heal corrupt npx cache in MCP stdio discovery: on ERR_MODULE_NOT_FOUND pointing into ~/.npm/_npx/<hash>/, wipe

that dir and retry once — fixes Discord/YouTube failing with "MCP stdio process exited unexpectedly" after partial npx installs,
  per-browser auto-close: spawned browser cards now disappear on their own sub-agent's completion instead of lingering until the parent agent
  finishes, WebSocketManager agent:status matches by session.browser_id (spawned_by is the parent, so it never matched the finishing
  sub-agent), user-created/pre-selected browsers untouched

9router/open-sse/translator/request/claude-to-openai.js

@@ -20,12 +20,21 @@ export function claudeToOpenAIRequest(model, body, stream) {
     result.temperature = body.temperature;
   }
 
-  // System message
+  // System message — strip the claude-agent-sdk's "You are Claude Code..."
+  // identity sentence before forwarding to non-Anthropic providers. Without
+  // this, Gemini/Codex/etc receive the Claude Code preset's first line as
+  // gospel and dutifully introduce themselves as Claude. The rest of the
+  // preset (tool conventions, scaffolding) stays intact.
   if (body.system) {
-    const systemContent = Array.isArray(body.system)
+    let systemContent = Array.isArray(body.system)
       ? body.system.map(s => s.text || "").join("\n")
       : body.system;
-    
+
+    systemContent = systemContent.replace(
+      /You are Claude Code, Anthropic's official CLI for Claude\.\s*/,
+      ""
+    );
+
     if (systemContent) {
       result.messages.push({
         role: "system",

backend/apps/tools_lib/tools_lib.py

@@ -716,8 +716,38 @@ async def _discover_mcp_tools_sse(url: str, headers: dict | None = None) -> list
         raise HTTPException(status_code=502, detail=f"SSE discovery failed: {first}") from first
 
 
-async def _discover_mcp_tools_stdio(command: str, args: list[str] | None = None, env: dict | None = None) -> list[dict]:
-    """Spawn a stdio MCP server process and call tools/list via JSON-RPC over stdin/stdout."""
+_NPX_CACHE_RE = re.compile(r"_npx[/\\]([0-9a-f]{8,})[/\\]")
+
+
+def _try_heal_npx_cache(stderr: str) -> str | None:
+    """On `ERR_MODULE_NOT_FOUND` pointing into `~/.npm/_npx/<hash>/`, wipe that one dir.
+
+    Why: interrupted npx installs leave a `package-lock.json` in the cache dir so
+    subsequent spawns reuse a partially-extracted node_modules tree, which dies at
+    import time. Scoped strictly to the extracted hash subdir — never touches
+    anything outside `~/.npm/_npx/`.
+    """
+    if "ERR_MODULE_NOT_FOUND" not in stderr:
+        return None
+    m = _NPX_CACHE_RE.search(stderr)
+    if not m:
+        return None
+    hash_ = m.group(1)
+    cache_dir = os.path.join(os.path.expanduser("~"), ".npm", "_npx", hash_)
+    if not os.path.isdir(cache_dir):
+        return None
+    logger.warning("Corrupted npx cache detected at %s; wiping and letting caller retry", cache_dir)
+    shutil.rmtree(cache_dir, ignore_errors=True)
+    return hash_
+
+
+async def _discover_mcp_tools_stdio(command: str, args: list[str] | None = None, env: dict | None = None, _attempt: int = 0) -> list[dict]:
+    """Spawn a stdio MCP server process and call tools/list via JSON-RPC over stdin/stdout.
+
+    On the first attempt, a failure that looks like corrupted npx cache
+    (`ERR_MODULE_NOT_FOUND` pointing into `~/.npm/_npx/<hash>/`) triggers one
+    auto-heal + retry. No heal on `_attempt >= 1`.
+    """
     cmd_path = _resolve_command(command)
     if not cmd_path:
         raise HTTPException(status_code=400, detail=f"Command '{command}' not found on PATH or common install locations")
@@ -782,7 +812,9 @@ async def _discover_mcp_tools_stdio(command: str, args: list[str] | None = None,
         tools_list = data.get("result", {}).get("tools", [])
         return [{"name": t.get("name", ""), "description": t.get("description", ""), "inputSchema": t.get("inputSchema")} for t in tools_list]
 
-    except HTTPException:
+    except HTTPException as e:
+        if _attempt == 0 and _try_heal_npx_cache(str(e.detail) if e.detail is not None else ""):
+            return await _discover_mcp_tools_stdio(command, args, env, _attempt=1)
         raise
     except asyncio.TimeoutError:
         raise HTTPException(status_code=504, detail="MCP stdio server timed out during discovery")

frontend/src/shared/ws/WebSocketManager.ts

@@ -129,17 +129,16 @@ class WebSocketManager {
         if (data.status === 'running' && session_id) {
           store.dispatch(trackAgentNotification(session_id));
         }
-        // Auto-remove browsers spawned by this agent when it reaches a
-        // terminal state on its own. agent:closed only fires when the user
-        // clicks X to close the session, so without this hook the browser
-        // cards would linger after natural completion or error. 'stopped'
-        // is intentionally skipped to preserve the "inspect after manual
-        // stop" affordance — matching the agent:closed branch below.
-        if (session_id && (data.status === 'completed' || data.status === 'error')) {
-          const browserCards = store.getState().dashboardLayout.browserCards;
-          for (const card of Object.values(browserCards)) {
-            if (card.spawned_by === session_id) {
-              store.dispatch(removeBrowserCard(card.browser_id));
+        // Per-sub-agent close via browser_id; skip user-created cards (no spawned_by).
+        if (
+          (data.status === 'completed' || data.status === 'error') &&
+          data.session?.mode === 'browser-agent'
+        ) {
+          const browserId = data.session.browser_id;
+          if (browserId) {
+            const card = store.getState().dashboardLayout.browserCards[browserId];
+            if (card && card.spawned_by) {
+              store.dispatch(removeBrowserCard(browserId));
             }
           }
         }