fix bugs

1. "first keyword: --"  cosmetic bug
2. re error

skills/pyspark-sql-guardrails/scripts/sql_guard.py

@@ -16,6 +16,34 @@ _FORBIDDEN_SQL = re.compile(
     re.IGNORECASE,
 )
 _LEADING_COMMENTS = re.compile(r"\A\s*(?:--[^\n]*\n|/\*.*?\*/\s*)*", re.DOTALL)
+_STRING_LITERAL = re.compile(r"'(?:[^']|'')*'|\"(?:[^\"]|\"\")*\"")
+
+
+def _strip_string_literals(sql: str) -> str:
+    """Replace quoted string literals with empty strings so that the
+    forbidden-keyword regex does not false-positive on text content
+    inside quotes (e.g. ``SELECT 'drop' AS note``).
+
+    Spark uses single quotes for strings and double quotes for delimited
+    identifiers; both are replaced because the forbidden-keyword set is
+    verb-shaped (DROP, DELETE, ...) and never a legitimate identifier.
+    The doubled-quote escape (``''`` / ``""``) is handled inside the regex.
+    """
+    return _STRING_LITERAL.sub("''", sql)
+
+
+def first_keyword(sql: str) -> str:
+    """Return the lowercased first real SQL keyword after stripping a
+    leading comment and a trailing semicolon. Pure inspection helper —
+    does not raise and does not enforce the allowlist.
+
+    Intended for CLI reporting so ``first keyword:`` reflects what the
+    validator actually saw, not the leading comment marker (``--``).
+    """
+    text = sql.strip()
+    body = text[:-1].strip() if text.endswith(";") else text
+    normalized = _LEADING_COMMENTS.sub("", body).lstrip()
+    return normalized.split(None, 1)[0].lower() if normalized else ""
 
 
 def assert_select_or_insert(sql: str) -> str:
@@ -38,7 +66,7 @@ def assert_select_or_insert(sql: str) -> str:
     if first not in {"select", "with", "insert"}:
         raise ValueError(f"Only SELECT and INSERT SQL are allowed, got {first!r}")
 
-    if _FORBIDDEN_SQL.search(normalized):
+    if _FORBIDDEN_SQL.search(_strip_string_literals(normalized)):
         raise ValueError("Forbidden SQL keyword found")
 
     if first == "with" and not re.search(r"\bselect\b", normalized, re.IGNORECASE):

skills/pyspark-sql-guardrails/scripts/test_sql_guard.py

@@ -44,6 +44,9 @@ expect_ok("trailing semicolon",  "SELECT 1 FROM dual;")
 expect_ok("leading -- comment",  "-- comment\nSELECT 1")
 expect_ok("leading /* comment",  "/* hi */ SELECT 1")
 expect_ok("INSERT ... SELECT",   "INSERT INTO t SELECT 1")
+expect_ok("string 'drop'",       "SELECT 'drop' AS note FROM dual")
+expect_ok("quoted identifier",   'SELECT "drop" AS note FROM dual')
+expect_ok("escaped quote",       "SELECT 'don''t drop' AS note FROM dual")
 
 print()
 print("=== EXPECT_RAISE ===")

skills/pyspark-sql-guardrails/scripts/validate_sql.py

@@ -32,7 +32,7 @@ import os
 SCRIPTS_DIR = os.path.dirname(os.path.abspath(__file__))
 sys.path.insert(0, SCRIPTS_DIR)
 
-from sql_guard import assert_select_or_insert  # noqa: E402
+from sql_guard import assert_select_or_insert, first_keyword  # noqa: E402
 
 
 def read_sql() -> str:
@@ -51,7 +51,7 @@ def main() -> int:
     except ValueError as e:
         print(f"FAIL - {e}")
         return 1
-    first = body.split(None, 1)[0].lower()
+    first = first_keyword(sql)
     print(f"PASS - first keyword: {first}, body length: {len(body)}")
     return 0