empathy-framework 4.7.1__py3-none-any.whl → 4.8.0__py3-none-any.whl

empathy_os/workflows/security_audit.py

@@ -342,11 +342,29 @@ class SecurityAuditWorkflow(BaseWorkflow):
                                     if self._is_detection_code(line_content, match.group()):
                                         continue
 
+                                    # Phase 2: Skip safe SQL parameterization patterns
+                                    if vuln_type == "sql_injection":
+                                        if self._is_safe_sql_parameterization(
+                                            line_content,
+                                            match.group(),
+                                            content,
+                                        ):
+                                            continue
+
                                     # Skip fake/test credentials
                                     if vuln_type == "hardcoded_secret":
                                         if self._is_fake_credential(match.group()):
                                             continue
 
+                                    # Phase 2: Skip safe random usage (tests, demos, documented)
+                                    if vuln_type == "insecure_random":
+                                        if self._is_safe_random_usage(
+                                            line_content,
+                                            file_name,
+                                            content,
+                                        ):
+                                            continue
+
                                     # Skip command_injection in documentation strings
                                     if vuln_type == "command_injection":
                                         if self._is_documentation_or_string(
@@ -380,6 +398,29 @@ class SecurityAuditWorkflow(BaseWorkflow):
                     except OSError:
                         continue
 
+        # Phase 3: Apply AST-based filtering for command injection
+        try:
+            from .security_audit_phase3 import apply_phase3_filtering
+
+            # Separate command injection findings
+            cmd_findings = [f for f in findings if f["type"] == "command_injection"]
+            other_findings = [f for f in findings if f["type"] != "command_injection"]
+
+            # Apply Phase 3 filtering to command injection
+            filtered_cmd = apply_phase3_filtering(cmd_findings)
+
+            # Combine back
+            findings = other_findings + filtered_cmd
+
+            logger.info(
+                f"Phase 3: Filtered command_injection from {len(cmd_findings)} to {len(filtered_cmd)} "
+                f"({len(cmd_findings) - len(filtered_cmd)} false positives removed)"
+            )
+        except ImportError:
+            logger.debug("Phase 3 module not available, skipping AST-based filtering")
+        except Exception as e:
+            logger.warning(f"Phase 3 filtering failed: {e}")
+
         input_tokens = len(str(input_data)) // 4
         output_tokens = len(str(findings)) // 4
 
@@ -541,6 +582,154 @@ class SecurityAuditWorkflow(BaseWorkflow):
 
         return False
 
+    def _is_safe_sql_parameterization(self, line_content: str, match_text: str, file_content: str) -> bool:
+        """Check if SQL query uses safe parameterization despite f-string usage.
+
+        Phase 2 Enhancement: Detects safe patterns like:
+        - placeholders = ",".join("?" * len(ids))
+        - cursor.execute(f"... IN ({placeholders})", ids)
+
+        This prevents false positives for the SQLite-recommended pattern
+        of building dynamic placeholder strings.
+
+        Args:
+            line_content: The line containing the match (may be incomplete for multi-line)
+            match_text: The matched text
+            file_content: Full file content for context analysis
+
+        Returns:
+            True if this is safe parameterized SQL, False otherwise
+        """
+        # Get the position of the match in the full file content
+        match_pos = file_content.find(match_text)
+        if match_pos == -1:
+            # Try to find cursor.execute
+            match_pos = file_content.find("cursor.execute")
+            if match_pos == -1:
+                return False
+
+        # Extract a larger context (next 200 chars after match)
+        context = file_content[match_pos:match_pos + 200]
+
+        # Also get lines before the match for placeholder detection
+        lines_before = file_content[:match_pos].split("\n")
+        recent_lines = lines_before[-10:] if len(lines_before) > 10 else lines_before
+
+        # Pattern 1: Check if this is a placeholder-based parameterized query
+        # Look for: cursor.execute(f"... IN ({placeholders})", params)
+        if "placeholders" in context or any("placeholders" in line for line in recent_lines[-5:]):
+            # Check if context has both f-string and separate parameters
+            # Pattern: f"...{placeholders}..." followed by comma and params
+            if re.search(r'f["\'][^"\']*\{placeholders\}[^"\']*["\']\s*,\s*\w+', context):
+                return True  # Safe - has separate parameters
+
+            # Also check if recent lines built the placeholders
+            for prev_line in reversed(recent_lines):
+                if "placeholders" in prev_line and '"?"' in prev_line and "join" in prev_line:
+                    # Found placeholder construction
+                    # Now check if the execute has separate parameters
+                    if "," in context and any(param in context for param in ["run_ids", "ids", "params", "values", ")"]):
+                        return True
+
+        # Pattern 2: Check if f-string only builds SQL structure with constants
+        # Example: f"SELECT * FROM {TABLE_NAME}" where TABLE_NAME is a constant
+        f_string_vars = re.findall(r'\{(\w+)\}', context)
+        if f_string_vars:
+            # Check if all variables are constants (UPPERCASE or table/column names)
+            all_constants = all(
+                var.isupper() or "TABLE" in var.upper() or "COLUMN" in var.upper()
+                for var in f_string_vars
+            )
+            if all_constants:
+                return True  # Safe - using constants, not user data
+
+        # Pattern 3: Check for security note comments nearby
+        # If developers added security notes, it's likely safe
+        for prev_line in reversed(recent_lines[-3:]):
+            if "security note" in prev_line.lower() and "safe" in prev_line.lower():
+                return True
+
+        return False
+
+    def _is_safe_random_usage(self, line_content: str, file_path: str, file_content: str) -> bool:
+        """Check if random usage is in a safe context (tests, simulations, non-crypto).
+
+        Phase 2 Enhancement: Reduces false positives for random module usage
+        in test fixtures, A/B testing simulations, and demo code.
+
+        Args:
+            line_content: The line containing the match
+            file_path: Path to the file being scanned
+            file_content: Full file content for context analysis
+
+        Returns:
+            True if random usage is safe/documented, False if potentially insecure
+        """
+        # Check if file is a test file
+        is_test = any(pattern in file_path.lower() for pattern in ["/test", "test_", "conftest"])
+
+        # Check for explicit security notes nearby
+        lines = file_content.split("\n")
+        line_index = None
+        for i, line in enumerate(lines):
+            if line_content.strip() in line:
+                line_index = i
+                break
+
+        if line_index is not None:
+            # Check 5 lines before and after for security notes
+            context_start = max(0, line_index - 5)
+            context_end = min(len(lines), line_index + 5)
+            context = "\n".join(lines[context_start:context_end]).lower()
+
+            # Look for clarifying comments
+            safe_indicators = [
+                "security note",
+                "not cryptographic",
+                "not for crypto",
+                "test data",
+                "demo data",
+                "simulation",
+                "reproducible",
+                "deterministic",
+                "fixed seed",
+                "not used for security",
+                "not used for secrets",
+                "not used for tokens",
+            ]
+
+            if any(indicator in context for indicator in safe_indicators):
+                return True  # Documented as safe
+
+        # Check for common safe random patterns
+        line_lower = line_content.lower()
+
+        # Pattern 1: Fixed seed (reproducible tests)
+        if "random.seed(" in line_lower:
+            return True  # Fixed seed is for reproducibility, not security
+
+        # Pattern 2: A/B testing, simulations, demos
+        safe_contexts = [
+            "simulation",
+            "demo",
+            "a/b test",
+            "ab_test",
+            "fixture",
+            "mock",
+            "example",
+            "sample",
+        ]
+        if any(context in file_path.lower() for context in safe_contexts):
+            return True
+
+        # If it's a test file without crypto indicators, it's probably safe
+        if is_test:
+            crypto_indicators = ["password", "secret", "token", "key", "crypto", "auth"]
+            if not any(indicator in file_path.lower() for indicator in crypto_indicators):
+                return True
+
+        return False
+
     async def _assess(self, input_data: dict, tier: ModelTier) -> tuple[dict, int, int]:
         """Risk scoring and severity classification.
 
@@ -674,6 +863,7 @@ class SecurityAuditWorkflow(BaseWorkflow):
         """
         try:
             from .security_adapters import _check_crew_available
+
             adapters_available = True
         except ImportError:
             adapters_available = False

empathy_os/workflows/security_audit_phase3.py

@@ -0,0 +1,328 @@
+"""Phase 3 Scanner Improvements - AST-based Command Injection Detection
+
+This module provides AST-based analysis for detecting actual eval/exec usage
+vs mentions in comments, docstrings, and documentation.
+
+Created: 2026-01-26
+Related: docs/SECURITY_PHASE2_COMPLETE.md
+"""
+
+import ast
+import logging
+from pathlib import Path
+from typing import Any
+
+logger = logging.getLogger(__name__)
+
+
+class EvalExecDetector(ast.NodeVisitor):
+    """AST visitor that detects actual eval() and exec() calls.
+
+    This visitor walks the AST to find real function calls to eval() and exec(),
+    distinguishing them from:
+    - String literals mentioning eval/exec
+    - Comments mentioning eval/exec
+    - Docstrings documenting security policies
+    - Detection code checking for eval/exec patterns
+    """
+
+    def __init__(self, file_path: str):
+        """Initialize detector.
+
+        Args:
+            file_path: Path to file being analyzed (for context)
+        """
+        self.file_path = file_path
+        self.findings: list[dict[str, Any]] = []
+        self._in_docstring = False
+        self._current_function = None
+
+    def visit_FunctionDef(self, node: ast.FunctionDef) -> None:
+        """Visit function definition to track context."""
+        self._current_function = node.name
+        self.generic_visit(node)
+        self._current_function = None
+
+    def visit_Call(self, node: ast.Call) -> None:
+        """Visit function call nodes to detect eval/exec."""
+        # Check if this is a call to eval() or exec()
+        func_name = None
+
+        if isinstance(node.func, ast.Name):
+            func_name = node.func.id
+        elif isinstance(node.func, ast.Attribute):
+            # Handle attribute access like obj.exec()
+            func_name = node.func.attr
+
+        if func_name in ("eval", "exec"):
+            # Found a real eval/exec call!
+            self.findings.append({
+                "type": "command_injection",
+                "function": func_name,
+                "line": node.lineno,
+                "col": node.col_offset,
+                "context": self._current_function,
+            })
+
+        self.generic_visit(node)
+
+
+def analyze_file_for_eval_exec(file_path: str | Path) -> list[dict[str, Any]]:
+    """Analyze a Python file for actual eval/exec usage using AST.
+
+    Args:
+        file_path: Path to Python file to analyze
+
+    Returns:
+        List of findings (actual eval/exec calls)
+
+    Example:
+        >>> findings = analyze_file_for_eval_exec("myfile.py")
+        >>> for finding in findings:
+        ...     print(f"{finding['function']} at line {finding['line']}")
+    """
+    file_path = Path(file_path)
+
+    if not file_path.exists():
+        return []
+
+    try:
+        content = file_path.read_text(encoding="utf-8", errors="ignore")
+        tree = ast.parse(content, filename=str(file_path))
+
+        detector = EvalExecDetector(str(file_path))
+        detector.visit(tree)
+
+        return detector.findings
+
+    except SyntaxError as e:
+        logger.debug(f"Syntax error parsing {file_path}: {e}")
+        return []
+    except Exception as e:
+        logger.debug(f"Error analyzing {file_path}: {e}")
+        return []
+
+
+def is_scanner_implementation_file(file_path: str) -> bool:
+    """Check if file is part of security scanner implementation.
+
+    Scanner files legitimately contain eval/exec patterns for detection
+    purposes and should not be flagged.
+
+    Args:
+        file_path: Path to check
+
+    Returns:
+        True if this is a scanner implementation file
+    """
+    scanner_indicators = [
+        # Scanner implementation files
+        "bug_predict",
+        "security_audit",
+        "security_scan",
+        "vulnerability_scan",
+        "owasp",
+        "secrets_detector",
+        "pii_scrubber",
+
+        # Pattern/rule definition files
+        "patterns.py",
+        "rules.py",
+        "checks.py",
+
+        # Test files for security scanners
+        "test_bug_predict",
+        "test_security",
+        "test_scanner",
+    ]
+
+    path_lower = file_path.lower()
+    return any(indicator in path_lower for indicator in scanner_indicators)
+
+
+def is_in_docstring_or_comment(line_content: str, file_content: str, line_num: int) -> bool:
+    """Enhanced check if line is in docstring or comment.
+
+    Phase 3 Enhancement: More robust detection of documentation context.
+
+    Args:
+        line_content: The line to check
+        file_content: Full file content
+        line_num: Line number (1-indexed)
+
+    Returns:
+        True if line is in docstring or comment
+    """
+    line = line_content.strip()
+
+    # Check for comment lines
+    if line.startswith("#"):
+        return True
+
+    # Check for inline comments
+    if "#" in line_content and line_content.index("#") < line_content.find("eval") if "eval" in line_content else True:
+        return True
+
+    # Parse file as AST to find docstrings
+    try:
+        tree = ast.parse(file_content)
+
+        # Get all docstrings
+        docstrings = []
+        for node in ast.walk(tree):
+            docstring = ast.get_docstring(node)
+            if docstring:
+                docstrings.append(docstring)
+
+        # Check if any docstring contains this line content
+        for docstring in docstrings:
+            if line_content.strip() in docstring:
+                return True
+
+    except SyntaxError:
+        pass
+
+    # Check for security policy patterns
+    security_patterns = [
+        "no eval",
+        "no exec",
+        "never use eval",
+        "never use exec",
+        "avoid eval",
+        "avoid exec",
+        "security:",
+        "- no eval",
+        "- no exec",
+    ]
+
+    line_lower = line.lower()
+    if any(pattern in line_lower for pattern in security_patterns):
+        return True
+
+    return False
+
+
+def enhanced_command_injection_detection(
+    file_path: str,
+    original_findings: list[dict[str, Any]]
+) -> list[dict[str, Any]]:
+    """Enhanced command injection detection with AST-based filtering.
+
+    Phase 3: Uses AST to distinguish actual eval/exec calls from mentions
+    in documentation, comments, and scanner implementation.
+
+    Args:
+        file_path: Path to file being analyzed
+        original_findings: Findings from regex-based detection
+
+    Returns:
+        Filtered list of actual vulnerabilities (not false positives)
+    """
+    # Step 1: Check if this is a scanner implementation file
+    if is_scanner_implementation_file(file_path):
+        return []  # Scanner files are allowed to mention eval/exec
+
+    # Step 2: For Python files, use AST-based detection
+    if file_path.endswith(".py"):
+        try:
+            ast_findings = analyze_file_for_eval_exec(file_path)
+
+            # Convert AST findings to format compatible with original
+            filtered = []
+            for finding in ast_findings:
+                filtered.append({
+                    "type": "command_injection",
+                    "file": file_path,
+                    "line": finding["line"],
+                    "match": f"{finding['function']}(",
+                    "severity": "critical",
+                    "owasp": "A03:2021 Injection",
+                    "context": finding.get("context", ""),
+                })
+
+            return filtered
+
+        except Exception as e:
+            logger.debug(f"AST analysis failed for {file_path}, falling back to regex: {e}")
+            # Fall back to original findings if AST fails
+            pass
+
+    # Step 3: For non-Python files or if AST fails, filter original findings
+    try:
+        file_content = Path(file_path).read_text(encoding="utf-8", errors="ignore")
+
+        filtered = []
+        for finding in original_findings:
+            line_num = finding.get("line", 0)
+            lines = file_content.split("\n")
+
+            if 0 < line_num <= len(lines):
+                line_content = lines[line_num - 1]
+
+                # Skip if in docstring or comment
+                if is_in_docstring_or_comment(line_content, file_content, line_num):
+                    continue
+
+                filtered.append(finding)
+
+        return filtered
+
+    except Exception as e:
+        logger.debug(f"Enhanced filtering failed for {file_path}: {e}")
+        return original_findings
+
+
+# =============================================================================
+# Integration with SecurityAuditWorkflow
+# =============================================================================
+
+
+def apply_phase3_filtering(findings: list[dict[str, Any]]) -> list[dict[str, Any]]:
+    """Apply Phase 3 AST-based filtering to command injection findings.
+
+    This is the main entry point for Phase 3 improvements.
+
+    Args:
+        findings: List of command injection findings from regex-based detection
+                 (should only contain command_injection type)
+
+    Returns:
+        Filtered list with false positives removed
+    """
+    if not findings:
+        return []
+
+    # Group findings by file
+    by_file: dict[str, list[dict[str, Any]]] = {}
+    for finding in findings:
+        file_path = finding.get("file", "")
+        if file_path not in by_file:
+            by_file[file_path] = []
+        by_file[file_path].append(finding)
+
+    # Apply enhanced detection per file
+    filtered_findings = []
+    for file_path, file_findings in by_file.items():
+        enhanced = enhanced_command_injection_detection(file_path, file_findings)
+        filtered_findings.extend(enhanced)
+
+    return filtered_findings
+
+
+if __name__ == "__main__":
+    # Test on known files
+    test_files = [
+        "src/empathy_os/workflows/bug_predict.py",
+        "src/empathy_os/orchestration/execution_strategies.py",
+        "tests/test_bug_predict_workflow.py",
+    ]
+
+    for file in test_files:
+        if Path(file).exists():
+            findings = analyze_file_for_eval_exec(file)
+            print(f"\n{file}:")
+            print(f"  Actual eval/exec calls: {len(findings)}")
+            for f in findings:
+                print(f"    Line {f['line']}: {f['function']}() in {f.get('context', 'module')}")
+        else:
+            print(f"\n{file}: Not found")