fix(encoding): read kernel pseudo-files as plain utf-8

utf-8-sig exists to tolerate BOMs that Windows tooling adds to files
users edit. /proc and /sys files are generated by the Linux kernel, never
BOM'd and absent on Windows, so -sig there only muddies the read/write
policy. Switch every literal /proc/ and /sys/ read to utf-8 and teach
the footgun read rule that string literals starting with /proc/ or
/sys/ are exempt (user-edited files keep utf-8-sig).

agent/proxy_sources/iron_proxy.py

@@ -584,7 +584,7 @@ def _read_pid() -> Optional[int]:
 def _pid_proc_starttime(pid: int) -> Optional[str]:
     """/proc/<pid>/stat starttime (field 22) on Linux, else None — cheap PID-recycling detector."""
     try:
-        text = Path(f"/proc/{pid}/stat").read_text(encoding="utf-8-sig")
+        text = Path(f"/proc/{pid}/stat").read_text(encoding="utf-8")
     except OSError:
         return None
     # comm may contain spaces/parens, so split after the LAST ")"; field 22 -> tail index 19.

gateway/cgroup_cleanup.py

@@ -21,7 +21,7 @@ from pathlib import Path
 def _own_cgroup_path() -> str | None:
     """Return the cgroup v2 path for the calling process, or None."""
     try:
-        text = Path("/proc/self/cgroup").read_text(encoding="utf-8-sig")
+        text = Path("/proc/self/cgroup").read_text(encoding="utf-8")
     except OSError:
         return None
     match = re.search(r"^0::(.+)$", text, re.MULTILINE)
@@ -30,7 +30,7 @@ def _own_cgroup_path() -> str | None:
 
 def _read_cgroup_pids(cgroup_path: str) -> list[int]:
     try:
-        raw = Path(f"/sys/fs/cgroup{cgroup_path}/cgroup.procs").read_text(encoding="utf-8-sig")
+        raw = Path(f"/sys/fs/cgroup{cgroup_path}/cgroup.procs").read_text(encoding="utf-8")
     except OSError:
         return []
     pids: list[int] = []

gateway/drain_control.py

@@ -48,11 +48,11 @@ def current_instantiation_epoch() -> str:
     """
     boot_id = pid1_start = ""
     with contextlib.suppress(OSError):
-        boot_id = Path("/proc/sys/kernel/random/boot_id").read_text(encoding="utf-8-sig").strip()
+        boot_id = Path("/proc/sys/kernel/random/boot_id").read_text(encoding="utf-8").strip()
     with contextlib.suppress(OSError, IndexError):
         # "<pid> (<comm>) <state> ...": comm may contain spaces/parens, so split on the
         # LAST ')'. starttime is field 22 (1-indexed) = tail index 19.
-        pid1_start = Path("/proc/1/stat").read_text(encoding="utf-8-sig").rsplit(")", 1)[1].split()[19]
+        pid1_start = Path("/proc/1/stat").read_text(encoding="utf-8").rsplit(")", 1)[1].split()[19]
     return f"{boot_id}:{pid1_start}" if (boot_id or pid1_start) else ""
 
 

gateway/shutdown_forensics.py

@@ -41,7 +41,7 @@ def _signal_name(sig: Any) -> str:
 
 def _read_proc_field(pid: int, key: str) -> Optional[str]:
     """Read a single field from /proc/<pid>/status.  Linux only; None elsewhere."""
-    with contextlib.suppress(OSError), open(f"/proc/{pid}/status", encoding="utf-8-sig") as fh:
+    with contextlib.suppress(OSError), open(f"/proc/{pid}/status", encoding="utf-8") as fh:
         for line in fh:
             if line.startswith(key + ":"):
                 return line.split(":", 1)[1].strip()
@@ -210,7 +210,7 @@ def check_systemd_timing_alignment(
         return None  # Not running under systemd (or at least not directly)
     # /proc/self/cgroup: "0::/user.slice/.../hermes-gateway.service"
     unit_name: Optional[str] = None
-    with contextlib.suppress(OSError), open("/proc/self/cgroup", encoding="utf-8-sig") as fh:
+    with contextlib.suppress(OSError), open("/proc/self/cgroup", encoding="utf-8") as fh:
         for line in fh:
             parts = reversed(line.strip().split("/"))
             unit_name = next((p for p in parts if p.endswith(".service")), None)

gateway/status.py

@@ -896,7 +896,7 @@ def _pid_exists(pid: int) -> bool:
 def _posix_is_zombie(pid: int) -> bool:
     """Zombie via ``/proc/<pid>/stat`` field 3, or ``ps -o state=`` without /proc (macOS/BSD)."""
     try:
-        stat_fields = Path(f"/proc/{pid}/stat").read_text(encoding="utf-8-sig").split()
+        stat_fields = Path(f"/proc/{pid}/stat").read_text(encoding="utf-8").split()
         return len(stat_fields) > 2 and stat_fields[2] == "Z"
     except FileNotFoundError:
         with contextlib.suppress(Exception):
@@ -1515,7 +1515,7 @@ def _scoped_lock_record_is_stale(existing: dict[str, Any], existing_pid: Optiona
 def _process_is_stopped(pid: int) -> bool:
     """True for a stopped / tracing-stop state (T/t) in ``/proc/<pid>/status``."""
     with contextlib.suppress(OSError):
-        for line in Path(f"/proc/{pid}/status").read_text(encoding="utf-8-sig").splitlines():
+        for line in Path(f"/proc/{pid}/status").read_text(encoding="utf-8").splitlines():
             if line.startswith("State:"):
                 return line.split()[1] in {"T", "t"}
     return False

hermes_cli/kanban_db_dispatch.py

@@ -328,7 +328,7 @@ def _pid_alive(pid: Optional[int]) -> bool:
         return False
     if sys.platform == "linux":
         try:
-            with open(f"/proc/{int(pid)}/status", "r", encoding="utf-8-sig") as f:
+            with open(f"/proc/{int(pid)}/status", "r", encoding="utf-8") as f:
                 for line in f:
                     if line.startswith("State:"):
                         # "State:\tZ (zombie)" → dead

hermes_cli/local_runtime/hardware.py

@@ -67,7 +67,7 @@ _LINUX_RAM_KEYS = frozenset({"MemTotal", "MemAvailable", "MemFree"})
 def _linux_meminfo_text() -> str | None:
     """Raw /proc/meminfo, or None when procfs cannot be read."""
     try:
-        return Path("/proc/meminfo").read_text(encoding="utf-8-sig")
+        return Path("/proc/meminfo").read_text(encoding="utf-8")
     except OSError:
         return None
 

hermes_cli/main_desktop.py

@@ -979,7 +979,7 @@ def _desktop_linux_needs_no_sandbox() -> bool:
     if hasattr(os, "geteuid") and os.geteuid() == 0:
         return False
     try:
-        with open("/proc/sys/kernel/apparmor_restrict_unprivileged_userns", encoding="utf-8-sig") as f:
+        with open("/proc/sys/kernel/apparmor_restrict_unprivileged_userns", encoding="utf-8") as f:
             return f.read().strip() == "1"
     except OSError:
         return False

hermes_cli/mem_trim.py

@@ -77,7 +77,7 @@ def _read_proc_status() -> str | None:
     if sys.platform != "linux":
         return None
     try:
-        return Path("/proc/self/status").read_text(encoding="utf-8-sig")
+        return Path("/proc/self/status").read_text(encoding="utf-8")
     except OSError:
         return None
 

hermes_cli/security_audit_startup.py

@@ -74,7 +74,7 @@ def _in_container() -> bool:
     if os.environ.get("HERMES_DESKTOP_CHILD_PID"):
         return False  # desktop child, not a server container
     try:
-        cgroup = Path("/proc/1/cgroup").read_text(encoding="utf-8-sig", errors="replace")
+        cgroup = Path("/proc/1/cgroup").read_text(encoding="utf-8", errors="replace")
     except Exception:
         return False
     return any(tok in cgroup for tok in ("docker", "containerd", "kubepods", "libpod"))
@@ -89,7 +89,7 @@ def _path_is_mounted(path: Path) -> bool:
     except Exception:
         target = path
     try:
-        mounts = Path("/proc/mounts").read_text(encoding="utf-8-sig", errors="replace").splitlines()
+        mounts = Path("/proc/mounts").read_text(encoding="utf-8", errors="replace").splitlines()
     except Exception:
         return True  # can't tell — fail safe (no warning)
     # (mountpoint, fstype) entries at or above target; the longest wins, first one on ties.

hermes_cli/web_server_lifecycle.py

@@ -28,7 +28,7 @@ def _process_start_marker(pid: int) -> str:
     """
     if sys.platform == "linux":
         try:
-            stat_line = Path(f"/proc/{pid}/stat").read_text(encoding="utf-8-sig")
+            stat_line = Path(f"/proc/{pid}/stat").read_text(encoding="utf-8")
         except FileNotFoundError as exc:
             raise ProcessLookupError(pid) from exc
 

scripts/check-windows-footguns.py

@@ -464,7 +464,10 @@ FOOTGUNS: list[Footgun] = [
             "path.read_text(encoding='utf-8-sig') / "
             "open(path, 'r', encoding='utf-8-sig')"
         ),
-        post_filter=lambda m, line: _is_read_shaped(line),
+        # Literal /proc/ and /sys/ paths are kernel pseudo-files: Linux
+        # generates them, no Windows tool can BOM them, and they do not
+        # exist on Windows at all, so plain utf-8 is the honest encoding.
+        post_filter=lambda m, line: _is_read_shaped(line) and not _KERNEL_PSEUDO_FILE.search(line),
     ),
     Footgun(
         name="write with encoding='utf-8-sig' (emits a BOM)",
@@ -696,6 +699,10 @@ def _extract_mode(line: str) -> str | None:
     return None
 
 
+# A string literal (plain or f-string) that starts with /proc/ or /sys/.
+_KERNEL_PSEUDO_FILE = re.compile(r"""['"]/(?:proc|sys)/""")
+
+
 def _is_read_shaped(line: str) -> bool:
     """Heuristic: does this line READ a file (so utf-8-sig applies)?
 

tests/scripts/test_footgun_encoding_direction.py

@@ -28,6 +28,9 @@ def linter():
     ('data = path.read_text(encoding="utf_8")', [(1, READ)]),
     ("with open(path, mode='r', encoding='utf-8') as f:", [(1, READ)]),
     ('path.read_text(encoding="utf-8-sig")', []),
+    ('Path(f"/proc/{pid}/stat").read_text(encoding="utf-8")', []),
+    ("with open('/sys/fs/cgroup/cgroup.procs', encoding='utf-8') as f:", []),
+    ('Path(proc_root, "stat").read_text(encoding="utf-8")  # /proc/ in prose only', [(1, READ)]),
     ('path.write_text(data, encoding="utf-8")', []),
     ("open(path, 'w', encoding='utf-8')", []),
     ("open(path, 'a', encoding='utf-8')", []),

tests/tools/test_bot_desktop_install.py

@@ -89,7 +89,7 @@ def test_timeout_kills_the_package_managers_whole_process_group(monkeypatch):
 
     def gone() -> bool:  # /proc-based: a reparented orphan sits outside our subtree, where os.kill(pid, 0) is guarded
         try:
-            return "Z" in (Path(f"/proc/{child}/stat").read_text(encoding="utf-8-sig").rsplit(")", 1)[1].split() or ["Z"])[0]
+            return "Z" in (Path(f"/proc/{child}/stat").read_text(encoding="utf-8").rsplit(")", 1)[1].split() or ["Z"])[0]
         except OSError:
             return True
 

tools/bot_desktop/browser.py

@@ -148,7 +148,7 @@ def _launched_by_session(chromium_pid: int) -> Optional[str]:
     for a human-started (dock) instance. Chromium itself gets a scrubbed environment, so the daemon's
     ``/proc/<ppid>/environ`` is the marker (Linux-only, same user)."""
     try:
-        with open(f"/proc/{chromium_pid}/status", encoding="utf-8-sig") as fh:
+        with open(f"/proc/{chromium_pid}/status", encoding="utf-8") as fh:
             ppid = next((int(line.split()[1]) for line in fh if line.startswith("PPid:")), 0)
         with open(f"/proc/{ppid}/environ", "rb") as fh:
             raw = fh.read()

tools/browser_tool_install.py

@@ -133,7 +133,7 @@ def _running_in_docker() -> bool:
     if os.path.exists("/.dockerenv"):
         return True
     try:
-        with open("/proc/1/cgroup", "rt", encoding="utf-8-sig") as fp:
+        with open("/proc/1/cgroup", "rt", encoding="utf-8") as fp:
             return "docker" in fp.read()
     except OSError:
         return False

tools/browser_tool_session.py

@@ -41,7 +41,7 @@ def apparmor_restricts_unprivileged_userns() -> bool:
     """Ubuntu 23.10+ default: unprivileged user namespaces are denied, so a Chromium whose
     ``chrome_sandbox`` helper is not setuid (Playwright's bundle) dies with 'No usable sandbox'."""
     try:
-        with open("/proc/sys/kernel/apparmor_restrict_unprivileged_userns", encoding="utf-8-sig") as f:
+        with open("/proc/sys/kernel/apparmor_restrict_unprivileged_userns", encoding="utf-8") as f:
             return f.read().strip() == "1"
     except OSError:
         return False

tools/process_registry.py

@@ -135,7 +135,7 @@ def _worker_memory_max_bytes() -> int:
                 override, _MIN_WORKER_MEMORY_MAX_BYTES // (1024 * 1024))
     candidates: List[int] = []
     try:
-        for line in Path("/proc/self/cgroup").read_text(encoding="utf-8-sig").splitlines():
+        for line in Path("/proc/self/cgroup").read_text(encoding="utf-8").splitlines():
             if line.startswith("0::"):
                 relative = line.partition("::")[2].lstrip("/")
                 raw_limit = (