Boot Chrome on demand instead of keeping it always running

CHANGELOG.md

@@ -2,6 +2,12 @@
 
 All notable changes to this project will be documented in this file.
 
+## [1.0.0.10] - 2026-07-07
+
+### Changed
+- Chrome now boots on demand: LinkedIn tool calls ping the CDP endpoint before attaching and launch it via `bin/browser-service start` if it's not up, waiting for CDP to come up. Session cookies persist on disk in the Chrome profile, so launchd/systemd no longer need to keep Chrome alive between calls.
+- `cron.system_status` no longer logs a startup warning when Chrome's CDP endpoint is unreachable — that's now the expected steady state — and status output/JSON drop the stale "reinstall the always-on service" restart hint in favor of noting it starts on demand.
+
 ## [1.0.0.9] - 2026-07-05
 
 ### Fixed

VERSION

@@ -1 +1 @@
-1.0.0.9
+1.0.0.10

cron/system_status.py

@@ -71,9 +71,6 @@ def probe_browser() -> dict[str, Any]:
     except (json.JSONDecodeError, TimeoutError, OSError) as exc:
         health_error = str(exc)
 
-    restart_hint = "bin/browser-service install"
-    if not managed:
-        restart_hint = "bin/browser-service install  (or ./install.sh)"
     return {
         "url": url,
         "reachable": reachable,
@@ -85,7 +82,7 @@ def probe_browser() -> dict[str, Any]:
         "service_backend": backend,
         "service_unit_path": str(unit_path) if unit_path else None,
         "auto_start_on_reboot": managed,
-        "restart_hint": restart_hint,
+        "starts_on_demand": True,
     }
 
 
@@ -95,15 +92,14 @@ def check_services(*, logger: logging.Logger | None = None) -> dict[str, Any]:
     whichever is unreachable. Intended to run alongside the startup
     "auto update check" — no email/paging, just a log line an operator
     tailing server.log will see.
+
+    Chrome now boots on demand (see outreach/browser.py:_ensure_browser_running),
+    so it being down at rest is the expected steady state, not a health issue —
+    only the cron server warrants a startup warning here.
     """
     log = logger or logging.getLogger(__name__)
     browser = probe_browser()
     cron_server = probe_cron_server()
-    if not browser.get("running"):
-        log.warning(
-            "Health check: browser CDP unreachable — %s",
-            browser.get("health_error") or "no response",
-        )
     if not cron_server.get("running"):
         log.warning(
             "Health check: cron server unreachable — %s",
@@ -124,10 +120,8 @@ def format_browser_lines() -> list[str]:
         lines.append(f"    profile    {browser.get('profile')}")
     else:
         err = browser.get("health_error")
-        hint = browser.get("restart_hint", "bin/browser-service install")
         extra = f" — {err}" if err else ""
-        lines.append(f"  Chrome CDP  not running{extra}")
-        lines.append(f"    restart: {hint}")
+        lines.append(f"  Chrome CDP  not running{extra}  (starts on demand)")
     if browser.get("managed"):
         lines.append(
             f"    auto-start  {browser.get('service_backend')}  ({browser.get('service_unit_path')})"

outreach/browser.py

@@ -537,6 +537,44 @@ def _pp_structure_activity_update(index: int, post: dict[str, Any]) -> dict[str,
 _ZERO_TAB_CDP_ERROR = "setDownloadBehavior"
 
 
+_REPO_ROOT = Path(__file__).resolve().parent.parent
+_BROWSER_SERVICE = _REPO_ROOT / "bin" / "browser-service"
+
+
+def _cdp_reachable(cdp_url: str, timeout: float = 2.0) -> bool:
+    try:
+        urllib.request.urlopen(f"{cdp_url}/json/version", timeout=timeout).close()
+        return True
+    except Exception:
+        return False
+
+
+async def _ensure_browser_running(cdp_url: str) -> None:
+    """
+    Boot Chrome on demand via bin/browser-service if the CDP endpoint isn't
+    reachable — so nothing needs to keep it running between tool calls.
+    browser-service already waits for CDP to come up before returning.
+    """
+    if await asyncio.to_thread(_cdp_reachable, cdp_url):
+        return
+    if not _BROWSER_SERVICE.is_file():
+        return
+    logger.info("Chrome not reachable at %s — starting it on demand", cdp_url)
+    env = dict(os.environ)
+    port = urlparse(cdp_url).port
+    if port:
+        env["CDP_PORT"] = str(port)
+    proc = await asyncio.create_subprocess_exec(
+        str(_BROWSER_SERVICE),
+        "start",
+        cwd=str(_REPO_ROOT),
+        env=env,
+        stdout=asyncio.subprocess.DEVNULL,
+        stderr=asyncio.subprocess.DEVNULL,
+    )
+    await proc.wait()
+
+
 def _open_blank_tab(cdp_url: str) -> None:
     """Ask Chrome's CDP HTTP endpoint to open a tab (bypasses Playwright context creation).
 
@@ -694,6 +732,7 @@ async def _attach(self) -> None:
 
         We never close a tab we didn't open, so the user's browsing is undisturbed.
         """
+        await _ensure_browser_running(self.cdp_url)
         self._browser = await self._connect_with_zero_tab_retry()
         self._is_attached = True
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "ebase"
-version = "1.0.0.9"
+version = "1.0.0.10"
 description = "LinkedIn recruiting outreach that runs inside Claude Code"
 readme = "README.md"
 license = "MIT"

testing/tests/test_system_status.py

@@ -13,7 +13,7 @@ def test_probe_browser_unreachable() -> None:
     browser = ss.probe_browser()
     assert "reachable" in browser
     assert "managed" in browser
-    assert "restart_hint" in browser
+    assert "starts_on_demand" in browser
 
 
 def test_format_browser_lines() -> None:
@@ -31,11 +31,16 @@ def test_check_services_returns_both_probes() -> None:
 def test_check_services_logs_warning_for_unreachable(
     caplog: pytest.LogCaptureFixture,
 ) -> None:
+    # Chrome boots on demand now, so being down is expected and shouldn't warn —
+    # only the cron server's health is worth a startup warning here.
     with caplog.at_level(logging.WARNING):
         result = ss.check_services()
-    for name, key in (("browser", "browser"), ("cron_server", "cron server")):
-        if not result[name].get("running"):
-            assert any(
-                record.levelno == logging.WARNING and key in record.getMessage().lower()
-                for record in caplog.records
-            ), f"expected a warning logged for unreachable {name}"
+    assert not any(
+        record.levelno == logging.WARNING and "browser" in record.getMessage().lower()
+        for record in caplog.records
+    )
+    if not result["cron_server"].get("running"):
+        assert any(
+            record.levelno == logging.WARNING and "cron server" in record.getMessage().lower()
+            for record in caplog.records
+        ), "expected a warning logged for unreachable cron server"