[eric] ci: a test process never spawns a router (OSW_NEVER_SPAWN_ROUTER), the boot hook's sync npm install hung the suite on every router-less runner; the slot-size test locates its source on Windows (ENG-486)

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

.github/workflows/suites-matrix.yml

@@ -28,6 +28,7 @@ jobs:
     env:
       OSW_NEVER_KILL_ROUTER: '1'
       OSW_DISABLE_AUTO_RESUME: '1'
+      OSW_NEVER_SPAWN_ROUTER: '1'
       # The packaged app spawns its backend with PYTHONUTF8=1 (electron/main.js); the suite runs the way the product runs, or Windows' cp1252 default fails collection on the first source read.
       PYTHONUTF8: '1'
       CSC_IDENTITY_AUTO_DISCOVERY: 'false'

backend/apps/nine_router/process.py

@@ -414,10 +414,28 @@ def p_report_start_failure(reason: str, *, detail: str = "", **fields: Any) -> N
         logger.debug("9router start-failure diagnostic submit failed", exc_info=True)
 
 
+def spawn_held_because() -> str | None:
+    """A declared switch, never an incidental fact: the suite and CI set it, the product never does."""
+    if os.environ.get("OSW_NEVER_SPAWN_ROUTER") == "1":
+        return "OSW_NEVER_SPAWN_ROUTER=1 (a test run adopts a router or has none)"
+    return None
+
+
+p_spawn_hold_said = False
+
+
 async def ensure_running():
     """Start 9Router if not already running. Serialized so concurrent callers
     (the background auto-start + a dispatch-time ensure) can't double-spawn."""
-    global p_start_lock
+    global p_start_lock, p_spawn_hold_said
+    held = spawn_held_because()
+    if held:
+        # The boot-time start ran a sync `npm install 9router` (up to 300 s) INSIDE the event loop on a runner with no
+        # router, and the whole suite hung behind it (macOS CI, 2026-09-06). A held spawn says so once and never blocks.
+        if not p_spawn_hold_said:
+            p_spawn_hold_said = True
+            logger.warning(f"[9router] NOT starting the router because {held}; every lane that needs it is unserved in this process")
+        return
     if p_start_lock is None:
         p_start_lock = asyncio.Lock()
     async with p_start_lock:
@@ -566,7 +584,6 @@ async def p_ensure_running_impl():
             if p_process is not None and p_process.poll() is None:
                 logger.info("9Router already running (ours) on port %d", NINE_ROUTER_PORT)
                 return
-            import subprocess as p_sp
             try:
                 p_stale = stale_router_pids()
                 if p_stale and router_kill_held_because():

backend/tests/conftest.py

@@ -24,7 +24,6 @@ assert "backend.config.paths" not in sys.modules, "conftest must set OPENSWARM_D
 os.environ["OPENSWARM_DATA_ROOT"] = tempfile.mkdtemp(prefix="osw_test_data_")
 
 import asyncio
-from datetime import datetime, timezone
 from types import SimpleNamespace
 
 import pytest
@@ -48,6 +47,8 @@ def _isolate_browser_state(monkeypatch):
     monkeypatch.setenv("OSW_DISABLE_AUTO_RESUME", "1")
     # A suite run must never kill a router it did not start; that is the user's app (ENG-393).
     monkeypatch.setenv("OSW_NEVER_KILL_ROUTER", "1")
+    # ...and never START one: the boot hook ran `npm install 9router` inside the event loop on a router-less CI runner and hung the suite.
+    monkeypatch.setenv("OSW_NEVER_SPAWN_ROUTER", "1")
     monkeypatch.setenv("OSW_PRESTAGE", "0")
     monkeypatch.setenv("OSW_FASTREAD_HOP", "0")
     monkeypatch.setenv("OSW_PRELUDE_TRIM", "0")

backend/tests/test_router_spawn_hold.py

@@ -0,0 +1,36 @@
+"""A test process never starts a router: the boot hook's sync npm install hung the whole suite on a router-less runner."""
+import logging
+
+import pytest
+
+from backend.apps.nine_router import process
+
+
+@pytest.mark.asyncio
+async def test_a_held_spawn_never_reaches_the_installer_and_says_so_once(monkeypatch):
+    monkeypatch.setenv("OSW_NEVER_SPAWN_ROUTER", "1")
+    process.p_spawn_hold_said = False
+
+    async def boom():
+        raise AssertionError("the installer ran under a held spawn")
+
+    monkeypatch.setattr(process, "p_ensure_running_impl", boom)
+    seen = []
+    handler = logging.Handler()
+    handler.emit = lambda rec: seen.append(rec.getMessage())
+    process.logger.addHandler(handler)
+    try:
+        await process.ensure_running()
+        await process.ensure_running()
+    finally:
+        process.logger.removeHandler(handler)
+    said = [m for m in seen if "NOT starting the router" in m]
+    assert len(said) == 1, said
+
+
+def test_the_suite_and_ci_declare_the_hold():
+    import os
+    import pathlib
+    assert os.environ.get("OSW_NEVER_SPAWN_ROUTER") == "1", "conftest must set it"
+    ci = pathlib.Path(".github/workflows/suites-matrix.yml").read_text(encoding="utf-8")
+    assert "OSW_NEVER_SPAWN_ROUTER: '1'" in ci

frontend/src/app/pages/AgentChat/browserSlotSize.test.ts

@@ -85,9 +85,10 @@ test('the component CSS still matches the constants this file tests', async () =
   const url = await import('node:url');
   const here = url.fileURLToPath(new URL('.', import.meta.url));
   // The test runs from .test-build, so resolve the source next to it by name.
+  // Both separators: on Windows `here` is D:\...\.test-build\..., and a '/'-only replace found nothing (CI, 2026-09-06).
   const candidates = [
     here + 'AgentChat.tsx',
-    here.replace('/.test-build/', '/src/') + 'AgentChat.tsx',
+    here.replace(/([\\/])\.test-build([\\/])/, '$1src$2') + 'AgentChat.tsx',
   ];
   const path = candidates.find((p) => fs.existsSync(p));
   assert.ok(path, `could not locate AgentChat.tsx from ${here}`);