fix(termux): harden env-backed background jobs

tests/tools/test_process_registry.py

@@ -340,6 +340,67 @@ class TestSpawnEnvSanitization:
         assert f"{_HERMES_PROVIDER_ENV_FORCE_PREFIX}TELEGRAM_BOT_TOKEN" not in env
         assert env["PYTHONUNBUFFERED"] == "1"
 
+    def test_spawn_via_env_uses_backend_temp_dir_for_artifacts(self, registry):
+        class FakeEnv:
+            def __init__(self):
+                self.commands = []
+
+            def get_temp_dir(self):
+                return "/data/data/com.termux/files/usr/tmp"
+
+            def execute(self, command, timeout=None):
+                self.commands.append((command, timeout))
+                return {"output": "4321\n"}
+
+        env = FakeEnv()
+        fake_thread = MagicMock()
+
+        with patch("tools.process_registry.threading.Thread", return_value=fake_thread), \
+            patch.object(registry, "_write_checkpoint"):
+            session = registry.spawn_via_env(env, "echo hello")
+
+        bg_command = env.commands[0][0]
+        assert session.pid == 4321
+        assert "/data/data/com.termux/files/usr/tmp/hermes_bg_" in bg_command
+        assert ".exit" in bg_command
+        assert "rc=$?;" in bg_command
+        assert " > /tmp/hermes_bg_" not in bg_command
+        assert "cat /tmp/hermes_bg_" not in bg_command
+        fake_thread.start.assert_called_once()
+
+    def test_env_poller_quotes_temp_paths_with_spaces(self, registry):
+        session = _make_session(sid="proc_space")
+        session.exited = False
+
+        class FakeEnv:
+            def __init__(self):
+                self.commands = []
+                self._responses = iter([
+                    {"output": "hello\n"},
+                    {"output": "1\n"},
+                    {"output": "0\n"},
+                ])
+
+            def execute(self, command, timeout=None):
+                self.commands.append((command, timeout))
+                return next(self._responses)
+
+        env = FakeEnv()
+
+        with patch("tools.process_registry.time.sleep", return_value=None), \
+            patch.object(registry, "_move_to_finished"):
+            registry._env_poller_loop(
+                session,
+                env,
+                "/path with spaces/hermes_bg.log",
+                "/path with spaces/hermes_bg.pid",
+                "/path with spaces/hermes_bg.exit",
+            )
+
+        assert env.commands[0][0] == "cat '/path with spaces/hermes_bg.log' 2>/dev/null"
+        assert env.commands[1][0] == "kill -0 \"$(cat '/path with spaces/hermes_bg.pid' 2>/dev/null)\" 2>/dev/null; echo $?"
+        assert env.commands[2][0] == "cat '/path with spaces/hermes_bg.exit' 2>/dev/null"
+
 
 # =========================================================================
 # Checkpoint

tools/process_registry.py

@@ -172,6 +172,19 @@ class ProcessRegistry:
 
     # ----- Spawn -----
 
+    @staticmethod
+    def _env_temp_dir(env: Any) -> str:
+        """Return the writable sandbox temp dir for env-backed background tasks."""
+        get_temp_dir = getattr(env, "get_temp_dir", None)
+        if callable(get_temp_dir):
+            try:
+                temp_dir = get_temp_dir()
+                if isinstance(temp_dir, str) and temp_dir.startswith("/"):
+                    return temp_dir.rstrip("/") or "/"
+            except Exception as exc:
+                logger.debug("Could not resolve environment temp dir: %s", exc)
+        return "/tmp"
+
     def spawn_local(
         self,
         command: str,
@@ -316,12 +329,20 @@ class ProcessRegistry:
         )
 
         # Run the command in the sandbox with output capture
-        log_path = f"/tmp/hermes_bg_{session.id}.log"
-        pid_path = f"/tmp/hermes_bg_{session.id}.pid"
+        temp_dir = self._env_temp_dir(env)
+        log_path = f"{temp_dir}/hermes_bg_{session.id}.log"
+        pid_path = f"{temp_dir}/hermes_bg_{session.id}.pid"
+        exit_path = f"{temp_dir}/hermes_bg_{session.id}.exit"
         quoted_command = shlex.quote(command)
+        quoted_temp_dir = shlex.quote(temp_dir)
+        quoted_log_path = shlex.quote(log_path)
+        quoted_pid_path = shlex.quote(pid_path)
+        quoted_exit_path = shlex.quote(exit_path)
         bg_command = (
-            f"nohup bash -c {quoted_command} > {log_path} 2>&1 & "
-            f"echo $! > {pid_path} && cat {pid_path}"
+            f"mkdir -p {quoted_temp_dir} && "
+            f"( nohup bash -lc {quoted_command} > {quoted_log_path} 2>&1; "
+            f"rc=$?; printf '%s\\n' \"$rc\" > {quoted_exit_path} ) & "
+            f"echo $! > {quoted_pid_path} && cat {quoted_pid_path}"
         )
 
         try:
@@ -342,7 +363,7 @@ class ProcessRegistry:
             # Start a poller thread that periodically reads the log file
             reader = threading.Thread(
                 target=self._env_poller_loop,
-                args=(session, env, log_path, pid_path),
+                args=(session, env, log_path, pid_path, exit_path),
                 daemon=True,
                 name=f"proc-poller-{session.id}",
             )
@@ -386,14 +407,17 @@ class ProcessRegistry:
         self._move_to_finished(session)
 
     def _env_poller_loop(
-        self, session: ProcessSession, env: Any, log_path: str, pid_path: str
+        self, session: ProcessSession, env: Any, log_path: str, pid_path: str, exit_path: str
     ):
         """Background thread: poll a sandbox log file for non-local backends."""
+        quoted_log_path = shlex.quote(log_path)
+        quoted_pid_path = shlex.quote(pid_path)
+        quoted_exit_path = shlex.quote(exit_path)
         while not session.exited:
             time.sleep(2)  # Poll every 2 seconds
             try:
                 # Read new output from the log file
-                result = env.execute(f"cat {log_path} 2>/dev/null", timeout=10)
+                result = env.execute(f"cat {quoted_log_path} 2>/dev/null", timeout=10)
                 new_output = result.get("output", "")
                 if new_output:
                     with session._lock:
@@ -403,14 +427,14 @@ class ProcessRegistry:
 
                 # Check if process is still running
                 check = env.execute(
-                    f"kill -0 $(cat {pid_path} 2>/dev/null) 2>/dev/null; echo $?",
+                    f"kill -0 \"$(cat {quoted_pid_path} 2>/dev/null)\" 2>/dev/null; echo $?",
                     timeout=5,
                 )
                 check_output = check.get("output", "").strip()
                 if check_output and check_output.splitlines()[-1].strip() != "0":
-                    # Process has exited -- get exit code
+                    # Process has exited -- get exit code captured by the wrapper shell.
                     exit_result = env.execute(
-                        f"wait $(cat {pid_path} 2>/dev/null) 2>/dev/null; echo $?",
+                        f"cat {quoted_exit_path} 2>/dev/null",
                         timeout=5,
                     )
                     exit_str = exit_result.get("output", "").strip()