feat(plugin): live AI guardrails — real-time rule violation interception

packages/claude-code-plugin/hooks/lib/rule_checker.py

@@ -0,0 +1,285 @@
+"""Live AI Guardrails — real-time rule violation detection (#1439).
+
+Intercepts Edit/Write tool calls and checks content against known
+security violation patterns (SQL injection, XSS, hardcoded secrets, etc.).
+Configurable via CODINGBUDDY_GUARDRAIL_LEVEL env var: strict/normal/off.
+"""
+import os
+import re
+from dataclasses import dataclass
+from typing import List, Optional
+
+
+@dataclass
+class Violation:
+    """A single rule violation found in code content."""
+
+    rule_id: str
+    severity: str
+    message: str
+    line_content: str
+    suggested_fix: str
+
+
+# --- Comment detection ---
+
+_COMMENT_RE = re.compile(r"^\s*(?:#|//|/\*|\*|<!--)")
+
+_VALID_LEVELS = {"strict", "normal", "off"}
+
+
+def _is_comment(line: str) -> bool:
+    """Check if a line is a comment."""
+    return bool(_COMMENT_RE.match(line))
+
+
+# --- Placeholder / false-positive filters for secrets ---
+
+_PLACEHOLDER_RE = re.compile(
+    r"(?:your[-_]|example|placeholder|changeme|xxx|test|dummy|fake|sample|TODO)",
+    re.IGNORECASE,
+)
+
+_ENV_REF_RE = re.compile(
+    r"(?:os\.environ|process\.env|getenv|ENV\[|config\.|settings\.)",
+    re.IGNORECASE,
+)
+
+
+# --- Pattern definitions ---
+# Each pattern: (compiled_regex, rule_id, severity, message, suggested_fix)
+# Patterns are applied per-line after filtering out comments.
+
+_SQL_KEYWORDS = r"(?:SELECT|INSERT|UPDATE|DELETE|DROP|ALTER|CREATE|MERGE|REPLACE)\b"
+
+# SQL injection: f-string, %-format, or concatenation with SQL keywords
+_SQL_FSTRING_RE = re.compile(
+    rf'f["\'].*{_SQL_KEYWORDS}.*\{{',
+    re.IGNORECASE,
+)
+_SQL_PERCENT_RE = re.compile(
+    rf'["\'].*{_SQL_KEYWORDS}.*%\s*(?:s|d|r)["\'].*%\s*\w',
+    re.IGNORECASE,
+)
+_SQL_CONCAT_RE = re.compile(
+    rf'["\'].*{_SQL_KEYWORDS}.*["\']\s*\+\s*\w',
+    re.IGNORECASE,
+)
+
+_NORMAL_PATTERNS = [
+    # SEC-001: SQL Injection
+    (
+        _SQL_FSTRING_RE,
+        "SEC-001",
+        "high",
+        "SQL injection detected — dynamic SQL via string interpolation",
+        "Use parameterized queries (e.g., cursor.execute('SELECT ... WHERE id = ?', (id,)))",
+    ),
+    (
+        _SQL_PERCENT_RE,
+        "SEC-001",
+        "high",
+        "SQL injection detected — dynamic SQL via %-formatting",
+        "Use parameterized queries instead of string formatting",
+    ),
+    (
+        _SQL_CONCAT_RE,
+        "SEC-001",
+        "high",
+        "SQL injection detected — dynamic SQL via string concatenation",
+        "Use parameterized queries instead of string concatenation",
+    ),
+    # SEC-002: XSS
+    (
+        re.compile(r"\.\s*innerHTML\s*=", re.IGNORECASE),
+        "SEC-002",
+        "high",
+        "XSS risk — direct innerHTML assignment",
+        "Use textContent for plain text, or sanitize with DOMPurify before assigning innerHTML",
+    ),
+    (
+        re.compile(r"document\.write\s*\("),
+        "SEC-002",
+        "high",
+        "XSS risk — document.write with potentially unsafe content",
+        "Use DOM APIs (createElement/textContent) instead of document.write",
+    ),
+    (
+        re.compile(r"dangerouslySetInnerHTML"),
+        "SEC-002",
+        "high",
+        "XSS risk — dangerouslySetInnerHTML bypasses React's XSS protection",
+        "Sanitize content with DOMPurify before using dangerouslySetInnerHTML",
+    ),
+    # SEC-004: Dangerous eval/exec
+    (
+        re.compile(r"\beval\s*\("),
+        "SEC-004",
+        "high",
+        "Dangerous eval() call — arbitrary code execution risk",
+        "Avoid eval(). Use JSON.parse() for data, or ast.literal_eval() for Python literals",
+    ),
+    (
+        re.compile(r"\bexec\s*\("),
+        "SEC-004",
+        "high",
+        "Dangerous exec() call — arbitrary code execution risk",
+        "Avoid exec(). Use safer alternatives like importlib or specific parsers",
+    ),
+]
+
+# SEC-003: Hardcoded secrets — handled separately with extra false-positive filtering
+_SECRET_KEY_RE = re.compile(
+    r"""(?:api[_-]?key|secret(?:[_-]?key)?|password|passwd|token|auth[_-]?token|private[_-]?key|access[_-]?key)"""
+    r"""\s*=\s*["']([^"']{8,})["']""",
+    re.IGNORECASE,
+)
+
+# AWS access key pattern
+_AWS_KEY_RE = re.compile(
+    r"""(?:aws[_-]?access[_-]?key(?:[_-]?id)?|aws[_-]?secret)\s*=\s*["']([A-Za-z0-9/+=]{16,})["']""",
+    re.IGNORECASE,
+)
+
+# Generic variable names containing "secret" but not caught above
+_GENERIC_SECRET_RE = re.compile(
+    r"""(?:secret|credential|private_key)\s*=\s*["']([^"']{8,})["']""",
+    re.IGNORECASE,
+)
+
+# Strict-only patterns
+_STRICT_PATTERNS = [
+    # SEC-005: Shell injection
+    (
+        re.compile(r"subprocess\.\w+\(.*shell\s*=\s*True", re.DOTALL),
+        "SEC-005",
+        "medium",
+        "Shell injection risk — subprocess with shell=True",
+        "Use subprocess.run(cmd_list) without shell=True, passing args as a list",
+    ),
+    (
+        re.compile(r"os\.system\s*\("),
+        "SEC-005",
+        "medium",
+        "Shell injection risk — os.system allows arbitrary command execution",
+        "Use subprocess.run(cmd_list) instead of os.system()",
+    ),
+]
+
+
+class RuleChecker:
+    """Checks code content against known security violation patterns."""
+
+    def __init__(self) -> None:
+        raw = os.environ.get("CODINGBUDDY_GUARDRAIL_LEVEL", "normal").lower()
+        self.level: str = raw if raw in _VALID_LEVELS else "normal"
+
+    def check(self, content: str) -> List[Violation]:
+        """Check code content for violations.
+
+        Args:
+            content: Code content string (possibly multiline).
+
+        Returns:
+            List of Violation objects found. Empty if level is 'off'.
+        """
+        if self.level == "off":
+            return []
+
+        violations: List[Violation] = []
+        lines = content.split("\n")
+
+        for line in lines:
+            stripped = line.strip()
+            if not stripped or _is_comment(stripped):
+                continue
+
+            # Normal patterns (always active)
+            for regex, rule_id, severity, message, fix in _NORMAL_PATTERNS:
+                if regex.search(stripped):
+                    violations.append(
+                        Violation(rule_id, severity, message, stripped, fix)
+                    )
+
+            # SEC-003: Hardcoded secrets (special handling)
+            self._check_secrets(stripped, violations)
+
+            # Strict-only patterns
+            if self.level == "strict":
+                for regex, rule_id, severity, message, fix in _STRICT_PATTERNS:
+                    if regex.search(stripped):
+                        violations.append(
+                            Violation(rule_id, severity, message, stripped, fix)
+                        )
+
+        return violations
+
+    def _check_secrets(self, line: str, violations: List[Violation]) -> None:
+        """Check a line for hardcoded secrets with false-positive filtering."""
+        # Skip if line references env vars
+        if _ENV_REF_RE.search(line):
+            return
+
+        # AWS access key IDs (AKIA prefix) — always flag, skip placeholder filter
+        aws_match = _AWS_KEY_RE.search(line)
+        if aws_match:
+            value = aws_match.group(1)
+            if len(value) >= 16:
+                violations.append(
+                    Violation(
+                        rule_id="SEC-003",
+                        severity="high",
+                        message="Hardcoded AWS credential detected — must not be in source code",
+                        line_content=line,
+                        suggested_fix="Use environment variables (AWS_ACCESS_KEY_ID) or AWS credentials file",
+                    )
+                )
+                return
+
+        for regex in (_SECRET_KEY_RE, _GENERIC_SECRET_RE):
+            match = regex.search(line)
+            if match:
+                value = match.group(1)
+                # Skip empty, very short, or placeholder values
+                if len(value) < 8:
+                    continue
+                if _PLACEHOLDER_RE.search(value):
+                    continue
+                violations.append(
+                    Violation(
+                        rule_id="SEC-003",
+                        severity="high",
+                        message="Hardcoded secret detected — credentials should not be in source code",
+                        line_content=line,
+                        suggested_fix="Use environment variables (os.environ / process.env) or a secrets manager",
+                    )
+                )
+                return  # One match per line is enough
+
+    def check_tool_input(
+        self, tool_name: str, tool_input: dict
+    ) -> List[Violation]:
+        """Check a PreToolUse tool_input for violations.
+
+        Only checks Edit and Write tool calls.
+
+        Args:
+            tool_name: Name of the tool being called.
+            tool_input: The tool_input dict from the hook.
+
+        Returns:
+            List of violations found in the content.
+        """
+        if tool_name not in ("Edit", "Write"):
+            return []
+
+        content: Optional[str] = None
+        if tool_name == "Edit":
+            content = tool_input.get("new_string", "")
+        elif tool_name == "Write":
+            content = tool_input.get("content", "")
+
+        if not content:
+            return []
+
+        return self.check(content)

packages/claude-code-plugin/hooks/lib/violation_renderer.py

@@ -0,0 +1,93 @@
+"""Violation message renderer for Live AI Guardrails (#1439).
+
+Renders clear, actionable violation messages with unicode box formatting,
+rule references, and suggested fixes.
+"""
+from typing import List
+
+from rule_checker import Violation
+
+
+# Severity display
+_SEVERITY_ICONS = {
+    "high": "\u2718",     # ✘
+    "medium": "\u26a0",   # ⚠
+    "low": "\u2139",      # ℹ
+}
+
+
+class ViolationRenderer:
+    """Renders violation messages for terminal and hook output."""
+
+    def render(self, violations: List[Violation]) -> str:
+        """Render violations as a formatted unicode box message.
+
+        Args:
+            violations: List of Violation objects.
+
+        Returns:
+            Formatted string with box-drawing characters, or empty string.
+        """
+        if not violations:
+            return ""
+
+        count = len(violations)
+        header = f" CodingBuddy Guardrail \u2502 {count} violation(s) found "
+        width = max(len(header) + 4, 60)
+
+        lines: list[str] = []
+        lines.append("\u250c" + "\u2500" * (width - 2) + "\u2510")
+        lines.append("\u2502" + header.center(width - 2) + "\u2502")
+        lines.append("\u251c" + "\u2500" * (width - 2) + "\u2524")
+
+        for i, v in enumerate(violations):
+            icon = _SEVERITY_ICONS.get(v.severity, "?")
+            lines.append(
+                "\u2502"
+                + f"  {icon} [{v.severity.upper()}] {v.rule_id}: {v.message}".ljust(width - 2)
+                + "\u2502"
+            )
+            # Truncate long line content
+            content = v.line_content
+            if len(content) > width - 12:
+                content = content[: width - 15] + "..."
+            lines.append(
+                "\u2502"
+                + f"    \u2514\u2500 {content}".ljust(width - 2)
+                + "\u2502"
+            )
+            lines.append(
+                "\u2502"
+                + f"    \u21b3 Fix: {v.suggested_fix}".ljust(width - 2)
+                + "\u2502"
+            )
+            if i < count - 1:
+                lines.append(
+                    "\u2502" + " " * (width - 2) + "\u2502"
+                )
+
+        lines.append("\u2514" + "\u2500" * (width - 2) + "\u2518")
+        return "\n".join(lines)
+
+    def render_for_hook(self, violations: List[Violation]) -> str:
+        """Render violations as concise additionalContext for PreToolUse hook.
+
+        Args:
+            violations: List of Violation objects.
+
+        Returns:
+            Compact string for hook additionalContext, or empty string.
+        """
+        if not violations:
+            return ""
+
+        parts: list[str] = []
+        parts.append(
+            f"[CodingBuddy Guardrail] {len(violations)} violation(s) detected:"
+        )
+        for v in violations:
+            parts.append(
+                f"  - {v.rule_id} ({v.severity.upper()}): {v.message}. "
+                f"Fix: {v.suggested_fix}"
+            )
+        return "\n".join(parts)