@misterhuydo/sentinel 1.6.13 → 1.6.15

package/.cairn/.hint-lock

@@ -1 +1 @@
-2026-04-27T17:46:15.506Z
+2026-04-28T09:31:55.053Z

package/.cairn/session.json

@@ -1,6 +1,6 @@
 {
-  "message": "Auto-checkpoint at 2026-04-27T17:47:43.806Z",
-  "checkpoint_at": "2026-04-27T17:47:43.808Z",
+  "message": "Auto-checkpoint at 2026-04-28T09:57:09.547Z",
+  "checkpoint_at": "2026-04-28T09:57:09.549Z",
   "active_files": [
     "J:\\Projects\\Sentinel\\cli\\bin\\sentinel.js",
     "J:\\Projects\\Sentinel\\cli\\lib\\test.js"

package/lib/generate.js

@@ -45,8 +45,15 @@ if [[ -f "$PID_FILE" ]] && kill -0 "$(cat "$PID_FILE")" 2>/dev/null; then
   exit 0
 fi
 
-# Kill any orphaned sentinel processes for this project (stale PIDs not in PID file)
-pkill -f "sentinel.main --config $DIR/config" 2>/dev/null || true
+# Kill any orphaned sentinel.main processes whose cwd is this project dir.
+# Match by /proc/PID/cwd (immune to relative-vs-absolute --config arg) so we
+# don't leak duplicate workers when watchdog spawns over a still-alive worker.
+for _pid in $(pgrep -f 'sentinel\\.main' 2>/dev/null); do
+  _pcwd=$(readlink -f "/proc/$_pid/cwd" 2>/dev/null) || continue
+  if [[ "$_pcwd" == "$DIR" ]]; then
+    kill "$_pid" 2>/dev/null && echo "[sentinel] killed orphaned sentinel-main PID $_pid (cwd=$DIR)"
+  fi
+done
 rm -f "$PID_FILE"
 
 WORKSPACE="$(dirname "$DIR")"
@@ -200,7 +207,12 @@ if [[ -f "$PID_FILE" ]] && kill -0 "$(cat "$PID_FILE")" 2>/dev/null; then
   echo "[sentinel] __NAME__ already running (PID $(cat "$PID_FILE"))"
   exit 0
 fi
-pkill -f "sentinel.main --config $DIR/config" 2>/dev/null || true
+for _pid in $(pgrep -f 'sentinel\.main' 2>/dev/null); do
+  _pcwd=$(readlink -f "/proc/$_pid/cwd" 2>/dev/null) || continue
+  if [[ "$_pcwd" == "$DIR" ]]; then
+    kill "$_pid" 2>/dev/null && echo "[sentinel] killed orphaned sentinel-main PID $_pid (cwd=$DIR)"
+  fi
+done
 rm -f "$PID_FILE"
 WORKSPACE="$(dirname "$DIR")"
 _claude_pro=true

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@misterhuydo/sentinel",
-  "version": "1.6.13",
+  "version": "1.6.15",
   "description": "Sentinel — Autonomous DevOps Agent installer and manager",
   "bin": {
     "sentinel": "./bin/sentinel.js"

package/python/sentinel/__init__.py

@@ -1 +1 @@
-__version__ = "1.6.13"
+__version__ = "1.6.15"

package/python/sentinel/config_loader.py

@@ -85,6 +85,7 @@ class SentinelConfig:
     auto_commit: bool = False        # project-level default: push directly to main (no PR); repos can override
     auto_release: bool = False       # project-level default: trigger CI/CD pipeline after push; repos can override
     auto_raise_issues: bool = False  # if True, auto-fix detected log errors without waiting for human to raise an issue
+    bot_watcher_noise_pattern: str = ""  # regex; matching bot-alert text is dropped before queueing as an issue
 
 
 @dataclass
@@ -284,6 +285,7 @@ class ConfigLoader:
             c.auto_release = d["AUTO_RELEASE"].lower() == "true"
         if "AUTO_RAISE_ISSUES" in d:
             c.auto_raise_issues = d["AUTO_RAISE_ISSUES"].lower() == "true"
+        c.bot_watcher_noise_pattern = d.get("BOT_WATCHER_NOISE_PATTERN", "").strip()
         self.sentinel = c
 
     def _load_log_sources(self):

package/python/sentinel/main.py

@@ -61,6 +61,20 @@ def _project_lock(project_name: str) -> "asyncio.Lock":
     return _project_locks[project_name]
 
 
+def _restart_via_execv() -> None:
+    """Re-exec the current process, preserving the original `python -m sentinel.main` invocation.
+
+    sys.argv[0] under -m mode is the absolute path to main.py — running that path
+    directly makes Python treat the file as a script, which breaks the relative
+    `from .cairn_client import ...` imports. Detect -m mode and re-launch with
+    `[python, '-m', 'sentinel.main', *original_args]` instead.
+    """
+    if __package__ and sys.argv and sys.argv[0].endswith(("main.py", "main")):
+        os.execv(sys.executable, [sys.executable, "-m", f"{__package__}.main", *sys.argv[1:]])
+    else:
+        os.execv(sys.executable, [sys.executable, *sys.argv])
+
+
 def _on_sigusr1(*_):
     global _report_requested
     _report_requested = True
@@ -619,6 +633,19 @@ async def _handle_issue(event: IssueEvent, cfg_loader: ConfigLoader, store: Stat
 async def _handle_issue_locked(event, repo, cfg_loader, store):
     """Heavy work portion of _handle_issue — serialised per project via _project_lock."""
     sentinel = cfg_loader.sentinel
+
+    # Re-check dedupe AFTER acquiring the lock — the pre-lock check in
+    # _handle_issue is racy when two poll cycles (or two worker processes
+    # sharing the same sentinel.db) both see the issue file before either has
+    # recorded an attempt. The lock then merely serialises duplicate work.
+    if store.fix_attempted_recently(event.fingerprint, hours=24):
+        logger.info(
+            "Issue %s skipped (post-lock recheck) — fingerprint %s already attempted",
+            event.source, event.fingerprint,
+        )
+        mark_done(event.issue_file)
+        return None
+
     auto_commit  = resolve_auto_commit(repo, sentinel)
     auto_release = resolve_auto_release(repo, sentinel)
 
@@ -945,7 +972,17 @@ async def _handle_issue_locked(event, repo, cfg_loader, store):
 
     except Exception:
         logger.exception("Unexpected error processing issue %s — archiving to prevent retry loop", event.source)
-        store.record_fix(event.fingerprint, "failed", repo_name=repo.repo_name)
+        # Don't insert a 'failed' row if the apply/publish flow already recorded
+        # an outcome (success or otherwise) for this fingerprint — otherwise a
+        # post-success exception (e.g. mark_done TOCTOU) would overwrite the
+        # success in Boss DM aggregation that picks the most recent row.
+        if not store.fix_attempted_recently(event.fingerprint, hours=24):
+            store.record_fix(event.fingerprint, "failed", repo_name=repo.repo_name)
+        else:
+            logger.info(
+                "Catch-all skipped recording 'failed' for %s — fixes table already has a recent row for this fingerprint",
+                event.fingerprint,
+            )
         mark_done(event.issue_file)
         return {"submitter": getattr(event, "submitter_user_id", ""),
                 "repo_name": repo.repo_name if repo else event.target_repo,
@@ -1481,7 +1518,7 @@ def _check_and_upgrade(cfg: SentinelConfig) -> bool:
     except Exception:
         pass
 
-    os.execv(sys.executable, [sys.executable] + sys.argv)
+    _restart_via_execv()
     return True  # unreachable after execv
 
 
@@ -1603,7 +1640,7 @@ async def _handle_dev_task(task, cfg_loader: ConfigLoader, store: StateStore):
             f"{mentions}:white_check_mark: *Patch finished* — running tests before restart...",
         )
         # Run test suite before restarting — revert if tests fail.
-        import os as _os, sys as _sys, subprocess as _sp
+        import subprocess as _sp
         _loop2 = asyncio.get_event_loop()
         _code_dir = Path(sentinel.sentinel_dev_repo_path or ".")
         _venv_pytest = _code_dir / ".venv" / "bin" / "pytest"
@@ -1666,7 +1703,7 @@ async def _handle_dev_task(task, cfg_loader: ConfigLoader, store: StateStore):
                 f"SOAK_MINUTES={_soak_mins}\n"
             )
             await asyncio.sleep(1)  # let the Slack message flush
-            _os.execv(_sys.executable, [_sys.executable] + _sys.argv)
+            _restart_via_execv()
     elif status == "needs_human":
         # Boss qualifies the raw Patch explanation before surfacing to users
         qualified = _boss_qualify_dev_reason(detail, sentinel)
@@ -2006,9 +2043,8 @@ async def _patch_soak_monitor(cfg_loader: ConfigLoader) -> None:
                      f":x: *Patch soak failed* — new errors detected after patch `{patch_hash[:8]}`. "
                      f"Reverted {'✓' if revert_ok else '(failed — check manually)'}. Restarting...")
         if revert_ok:
-            import os as _os, sys as _sys
             await asyncio.sleep(2)
-            _os.execv(_sys.executable, [_sys.executable] + _sys.argv)
+            _restart_via_execv()
         return
 
     # Clean soak — notify

package/python/sentinel/slack_bot.py

@@ -55,6 +55,30 @@ class _Session:
 _sessions: dict[str, _Session] = {}
 _sessions_lock = asyncio.Lock()
 
+# Cached compiled regex for bot-watcher noise filter — recompiled when the
+# config value changes (so SIGHUP reload picks up edits without a restart).
+_noise_filter_cache: tuple[str, "object | None"] = ("", None)
+
+
+def _get_noise_filter(pattern: str):
+    """Return a compiled re.Pattern for the noise filter, or None if disabled/invalid."""
+    global _noise_filter_cache
+    cached_pattern, cached_re = _noise_filter_cache
+    if pattern == cached_pattern:
+        return cached_re
+    if not pattern:
+        _noise_filter_cache = ("", None)
+        return None
+    import re as _re
+    try:
+        compiled = _re.compile(pattern, _re.IGNORECASE)
+    except _re.error as e:
+        logger.warning("BOT_WATCHER_NOISE_PATTERN is not a valid regex (%s) — filter disabled", e)
+        _noise_filter_cache = (pattern, None)
+        return None
+    _noise_filter_cache = (pattern, compiled)
+    return compiled
+
 
 async def _get_or_create_session(user_id: str, user_name: str, channel: str) -> _Session:
     async with _sessions_lock:
@@ -272,6 +296,23 @@ async def _handle_bot_message(event: dict, client, cfg_loader, store) -> None:
     if not text:
         return
 
+    # Drop operational-noise alerts (e.g. wrong-PIN attempts, single transient
+    # SMS failures) before they enter the issues queue. Configurable via
+    # BOT_WATCHER_NOISE_PATTERN in sentinel.properties (regex, alternatives
+    # joined with `|`). Empty = disabled.
+    noise_re = _get_noise_filter(getattr(cfg_loader.sentinel, "bot_watcher_noise_pattern", ""))
+    if noise_re is not None and noise_re.search(text):
+        logger.info(
+            "Bot watcher: dropping noise-pattern match from %s (preview: %r)",
+            bot_id, text[:120].replace("\n", " "),
+        )
+        if ts:
+            try:
+                await client.reactions_add(channel=channel, timestamp=ts, name="mute")
+            except Exception:
+                pass  # reactions:write scope may not be granted — non-fatal
+        return
+
     # Find the project this bot is registered to
     # Match on either the B-prefixed bot_id or the U-prefixed user ID
     _user_id = event.get("user", "")

package/python/tests/test_bot_noise_filter.py

@@ -0,0 +1,51 @@
+"""Tests for the bot-watcher noise-pattern filter (slack_bot._get_noise_filter)."""
+
+from sentinel import slack_bot
+
+
+def _reset_cache():
+    slack_bot._noise_filter_cache = ("", None)
+
+
+def test_empty_pattern_disables_filter():
+    _reset_cache()
+    assert slack_bot._get_noise_filter("") is None
+
+
+def test_invalid_regex_disables_filter(caplog):
+    _reset_cache()
+    import logging
+    with caplog.at_level(logging.WARNING):
+        assert slack_bot._get_noise_filter("(unbalanced") is None
+    assert any("not a valid regex" in r.message for r in caplog.records)
+
+
+def test_pattern_compiles_and_caches():
+    _reset_cache()
+    p1 = slack_bot._get_noise_filter("Illegal pin logon|SMS delivery FAILED")
+    assert p1 is not None
+    p2 = slack_bot._get_noise_filter("Illegal pin logon|SMS delivery FAILED")
+    assert p1 is p2  # cache hit returns the same compiled object
+
+
+def test_pattern_change_recompiles():
+    _reset_cache()
+    p1 = slack_bot._get_noise_filter("foo")
+    p2 = slack_bot._get_noise_filter("bar")
+    assert p1 is not p2
+
+
+def test_match_is_case_insensitive():
+    _reset_cache()
+    pat = slack_bot._get_noise_filter("Illegal pin logon")
+    assert pat.search("ILLEGAL PIN LOGON ATTEMPTED")
+    assert pat.search("user typed wrong PIN — Illegal Pin Logon attempted")
+    assert not pat.search("normal startup message")
+
+
+def test_alternatives_match_either():
+    _reset_cache()
+    pat = slack_bot._get_noise_filter("Illegal pin logon|1 SMS delivery FAILED")
+    assert pat.search("Illegal pin logon attempted.")
+    assert pat.search(":no_entry: 1 SMS delivery FAILED in the last 5 minutes")
+    assert not pat.search("MsGraphMailSender threw exception")

package/python/tests/test_config_loader.py

@@ -74,6 +74,21 @@ def test_sentinel_config(config_dir):
     assert cfg.github_token == "ghp_test"
 
 
+def test_bot_watcher_noise_pattern_default_empty(config_dir):
+    cfg = ConfigLoader(str(config_dir)).sentinel
+    assert cfg.bot_watcher_noise_pattern == ""
+
+
+def test_bot_watcher_noise_pattern_loaded(tmp_path):
+    (tmp_path / "log-configs").mkdir()
+    (tmp_path / "repo-configs").mkdir()
+    (tmp_path / "sentinel.properties").write_text(
+        "BOT_WATCHER_NOISE_PATTERN=Illegal pin logon|1 SMS delivery FAILED\n"
+    )
+    cfg = ConfigLoader(str(tmp_path)).sentinel
+    assert cfg.bot_watcher_noise_pattern == "Illegal pin logon|1 SMS delivery FAILED"
+
+
 def test_log_sources(config_dir):
     loader = ConfigLoader(str(config_dir))
     assert "elprint-salescore" in loader.log_sources