fix(agents): compare a claude process's lstart in UTC (Issue #31, Mac review)

time.mktime read ps's local lstart, which is ambiguous in the hour a DST
change repeats: the start could come out an hour off and a live holder be
taken for a reused pid (or the reverse). ps now runs with TZ=UTC0, which BSD
ps and procps both honour, and calendar.timegm reads it. No /proc.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

.agents/skills/lib_py/agents/adapters/claude.py

@@ -1,4 +1,4 @@
-import os, re, json, glob, math, stat, time, shutil, subprocess
+import os, re, json, glob, math, stat, time, shutil, calendar, subprocess
 from typing import Optional, Any
 from lib_py.agents.base import BaseAgentAdapter, DiscoveryContext, conversation_in_use, scan_names, _plausible_id
 from lib_py.verify_session import workspace_key
@@ -84,10 +84,11 @@ def _session_file_live(data: dict, pid) -> bool:
             or not math.isfinite(started) or started <= 0:
         return True
     started = started / 1000.0
+    # lstart in UTC: a local time is ambiguous in the hour a DST change repeats.
     try:
         out = subprocess.run(["ps", "-p", str(pid), "-o", "lstart="], capture_output=True, text=True,
-                             timeout=5, env=dict(os.environ, LC_ALL="C")).stdout
+                             timeout=5, env=dict(os.environ, LC_ALL="C", TZ="UTC0")).stdout
-        lstart = time.mktime(time.strptime(" ".join(out.split()), "%a %b %d %H:%M:%S %Y"))
+        lstart = calendar.timegm(time.strptime(" ".join(out.split()), "%a %b %d %H:%M:%S %Y"))
     except Exception:
         return True
     # startedAt is written 1.7-3.2 s after the process starts; 2 s covers

tests/test_issue31_purge_residual.py

@@ -447,6 +447,30 @@ def test_session_file_liveness_fails_safe(monkeypatch, live):
     assert claude_mod._session_file_live({"startedAt": hour_ago}, live.pid) is True, "ps unavailable: live"
 
 
+def test_liveness_reads_lstart_as_utc(tmp_path, monkeypatch, live):
+    """2026-11-01 01:30 happens twice in EST5EDT. A process that started at
+    the second one (06:30 UTC) shows lstart 01:30 to a local-time ps; read as
+    the first (05:30 UTC) it predates a session file written at 05:45. ps
+    runs with TZ=UTC0 instead, which BSD ps and procps both honour."""
+    import calendar
+    from lib_py.agents.adapters import claude as claude_mod
+    fake = tmp_path / "fakebin"
+    fake.mkdir()
+    (fake / "ps").write_text('#!/bin/sh\nif [ "$TZ" = UTC0 ]; then echo "Sun Nov  1 06:30:00 2026"; '
+                             'else echo "Sun Nov  1 01:30:00 2026"; fi\n')
+    (fake / "ps").chmod(0o755)
+    utc = lambda hhmmss: calendar.timegm(time.strptime(f"2026-11-01 {hhmmss}", "%Y-%m-%d %H:%M:%S")) * 1000  # noqa: E731
+    try:
+        with monkeypatch.context() as m:
+            m.setenv("PATH", f"{fake}:{os.environ['PATH']}")
+            m.setenv("TZ", "EST5EDT,M3.2.0,M11.1.0")
+            time.tzset()
+            assert claude_mod._session_file_live({"startedAt": utc("05:45:00")}, live.pid) is False, "reused pid"
+            assert claude_mod._session_file_live({"startedAt": utc("06:30:02")}, live.pid) is True
+    finally:
+        time.tzset()
+
+
 @pytest.mark.parametrize("started", ["NaN", float("nan"), float("inf"), float("-inf"), "inf", "1760000000000",
                                      "soon", True, False, None, [1], "<missing>"])
 def test_an_unusable_started_at_means_live(live, started):