Replace the simple substring-based looks_dangerous() function in run_shell_command
with comprehensive pattern matching via ShellSafetyService. The new implementation
checks commands against 15 regex-based patterns across four severity levels
(LOW, MEDIUM, HIGH, CRITICAL) before executing shell commands.

Key changes:
- Module-level ShellSafetyService instance used for safety gating
- looks_dangerous() preserved as backward-compatible function delegating to service
- Legacy confirm_dangerous(callback) parameter adapted via wrapper
- New warn_callback(warning) parameter for rich warning context
- Comprehensive BDD test scenarios for the integration

@@ -0,0 +43,4 @@
+# ---------------------------------------------------------------------------
+# Comprehensive danger detection (replaces legacy looks_dangerous)
+# ---------------------------------------------------------------------------
+

@@ -0,0 +146,4 @@
+
+# ---------------------------------------------------------------------------
+# New-style warn_callback with full warning details
+# ---------------------------------------------------------------------------

@@ -0,0 +168,4 @@
+)
+def step_run_dangerous_with_warn(context, command):
+    """Run a dangerous command using the new-style warn_callback parameter."""
+    fake_proc = subprocess.CompletedProcess(

@@ -33,0 +85,4 @@
+        return None
+
+    def _wrapper(warning: DangerousCommandWarning) -> bool:
+        return callback(warning.command)  # type: ignore[arg-type]