LLM-Driven Extraction of Unstructured Data — Built for API Deployments & ETL Pipeline Workflows
Python ai-agentsdata-engineeringdocument-aigenerative-aiidpjson-extractionllmmcp-serverocrpdf-extractionprompt-engineeringstructured-output
[SECURITY] Add input validation and CSP headers (#1834)
Hari John Kuriakosebf15e8d2e201
diff --git a/backend/account_v2/serializer.py b/backend/account_v2/serializer.pyindex 4011cf4d5..e189b4efe 100644--- a/backend/account_v2/serializer.py+++ b/backend/account_v2/serializer.py@@ -1,6 +1,7 @@import refrom rest_framework import serializers+from utils.input_sanitizer import validate_name_fieldfrom account_v2.models import Organization, User@@ -10,6 +11,12 @@ class OrganizationSignupSerializer(serializers.Serializer):display_name = serializers.CharField(required=True, max_length=150)organization_id = serializers.CharField(required=True, max_length=30)+ def validate_name(self, value: str) -> str:+ return validate_name_field(value, field_name="Organization name")++ def validate_display_name(self, value: str) -> str:+ return validate_name_field(value, field_name="Display name")+def validate_organization_id(self, value): # type: ignoreif not re.match(r"^[a-z0-9_-]+$", value):raise serializers.ValidationError(diff --git a/backend/adapter_processor_v2/serializers.py b/backend/adapter_processor_v2/serializers.pyindex 482a6a9e7..627731ed8 100644--- a/backend/adapter_processor_v2/serializers.py+++ b/backend/adapter_processor_v2/serializers.py@@ -6,6 +6,7 @@ from cryptography.fernet import Fernetfrom django.conf import settingsfrom rest_framework import serializersfrom rest_framework.serializers import ModelSerializer+from utils.input_sanitizer import validate_name_field, validate_no_html_tagsfrom adapter_processor_v2.adapter_processor import AdapterProcessorfrom adapter_processor_v2.constants import AdapterKeys@@ -28,6 +29,20 @@ class BaseAdapterSerializer(AuditSerializer):model = AdapterInstancefields = "__all__"+ def validate(self, data):+ data = super().validate(data)+ adapter_name = data.get("adapter_name")+ if adapter_name is not None:+ data["adapter_name"] = validate_name_field(+ adapter_name, field_name="Adapter name"+ )+ description = data.get("description")+ if description is not None:+ data["description"] = validate_no_html_tags(+ description, field_name="Description"+ )+ return data+class DefaultAdapterSerializer(serializers.Serializer):llm_default = serializers.CharField(max_length=FLC.UUID_LENGTH, required=False)diff --git a/backend/api_v2/serializers.py b/backend/api_v2/serializers.pyindex ef5bcb243..7c9a5a769 100644--- a/backend/api_v2/serializers.py+++ b/backend/api_v2/serializers.py@@ -22,6 +22,7 @@ from rest_framework.serializers import (ValidationError,)from tags.serializers import TagParamsSerializer+from utils.input_sanitizer import validate_name_field, validate_no_html_tagsfrom utils.serializer.integrity_error_mixin import IntegrityErrorMixinfrom workflow_manager.endpoint_v2.models import WorkflowEndpointfrom workflow_manager.workflow_v2.exceptions import ExecutionDoesNotExistError@@ -62,6 +63,14 @@ class APIDeploymentSerializer(IntegrityErrorMixin, AuditSerializer):api_name_validator(value)return value+ def validate_display_name(self, value: str) -> str:+ return validate_name_field(value, field_name="Display name")++ def validate_description(self, value: str) -> str:+ if value is None:+ return value+ return validate_no_html_tags(value, field_name="Description")+def validate_workflow(self, workflow):"""Validate that the workflow has properly configured source and destination endpoints."""# Get all endpoints for this workflow with related datadiff --git a/backend/backend/settings/base.py b/backend/backend/settings/base.pyindex b71467e43..bb62960ae 100644--- a/backend/backend/settings/base.py+++ b/backend/backend/settings/base.py@@ -419,6 +419,7 @@ MIDDLEWARE = ["social_django.middleware.SocialAuthExceptionMiddleware","middleware.remove_allow_header.RemoveAllowHeaderMiddleware","middleware.cache_control.CacheControlMiddleware",+ "middleware.content_security_policy.ContentSecurityPolicyMiddleware",]TENANT_SUBFOLDER_PREFIX = f"{PATH_PREFIX}/unstract"diff --git a/backend/connector_v2/serializers.py b/backend/connector_v2/serializers.pyindex 45d5c0756..5517bc525 100644--- a/backend/connector_v2/serializers.py+++ b/backend/connector_v2/serializers.py@@ -9,6 +9,7 @@ from connector_processor.constants import ConnectorKeysfrom connector_processor.exceptions import OAuthTimeOutfrom rest_framework.serializers import CharField, SerializerMethodFieldfrom utils.fields import EncryptedBinaryFieldSerializer+from utils.input_sanitizer import validate_name_fieldfrom backend.serializers import AuditSerializerfrom connector_v2.constants import ConnectorInstanceKey as CIKey@@ -28,6 +29,9 @@ class ConnectorInstanceSerializer(AuditSerializer):model = ConnectorInstancefields = "__all__"+ def validate_connector_name(self, value: str) -> str:+ return validate_name_field(value, field_name="Connector name")+def save(self, **kwargs): # type: ignoreuser = self.context.get("request").user or Noneconnector_id: str = kwargs[CIKey.CONNECTOR_ID]diff --git a/backend/middleware/content_security_policy.py b/backend/middleware/content_security_policy.pynew file mode 100644index 000000000..619bb10c0--- /dev/null+++ b/backend/middleware/content_security_policy.py@@ -0,0 +1,40 @@+from django.http import HttpRequest, HttpResponse+from django.utils.deprecation import MiddlewareMixin+++class ContentSecurityPolicyMiddleware(MiddlewareMixin):+ """Middleware to add Content-Security-Policy header to all responses.++ The policy is restrictive by default. The inline script in login.html+ is allowed via a SHA-256 hash rather than 'unsafe-inline' to maintain+ strong XSS protection. 'unsafe-inline' is only used for style-src+ because the login page uses inline <style> blocks.++ Uses response.setdefault() so that route-specific CSP policies set by+ views or earlier middleware are not overwritten.+ """++ # SHA-256 hash of the inline script in login.html (form submit spinner).+ # If that script changes, regenerate with:+ # python -c "import hashlib,base64; ..."+ _SCRIPT_HASH = "sha256-GES82NvXpRYmVFDKv6vRHx2c7xuv8mgUzUaP7heKeFY="++ def process_response(+ self, request: HttpRequest, response: HttpResponse+ ) -> HttpResponse:+ response.setdefault(+ "Content-Security-Policy",+ (+ "default-src 'self'; "+ f"script-src 'self' '{self._SCRIPT_HASH}'; "+ "style-src 'self' 'unsafe-inline'; "+ "img-src 'self'; "+ "font-src 'self'; "+ "connect-src 'self'; "+ "object-src 'none'; "+ "frame-ancestors 'none'; "+ "base-uri 'self'; "+ "form-action 'self'"+ ),+ )+ return responsediff --git a/backend/notification_v2/serializers.py b/backend/notification_v2/serializers.pyindex cd82fed8a..115487c48 100644--- a/backend/notification_v2/serializers.py+++ b/backend/notification_v2/serializers.py@@ -1,4 +1,5 @@from rest_framework import serializers+from utils.input_sanitizer import validate_name_fieldfrom .enums import AuthorizationType, NotificationType, PlatformTypefrom .models import Notification@@ -109,6 +110,8 @@ class NotificationSerializer(serializers.ModelSerializer):"""Check uniqueness of the name with respect to either 'api' or'pipeline'."""+ value = validate_name_field(value, field_name="Notification name")+api = self.initial_data.get("api", getattr(self.instance, "api", None))pipeline = self.initial_data.get("pipeline", getattr(self.instance, "pipeline", None)diff --git a/backend/prompt_studio/prompt_studio_core_v2/serializers.py b/backend/prompt_studio/prompt_studio_core_v2/serializers.pyindex c45367a95..14ce69be5 100644--- a/backend/prompt_studio/prompt_studio_core_v2/serializers.py+++ b/backend/prompt_studio/prompt_studio_core_v2/serializers.py@@ -8,6 +8,7 @@ from django.core.exceptions import ObjectDoesNotExistfrom rest_framework import serializersfrom rest_framework.exceptions import ValidationErrorfrom utils.FileValidator import FileValidator+from utils.input_sanitizer import validate_name_field, validate_no_html_tagsfrom utils.serializer.integrity_error_mixin import IntegrityErrorMixinfrom backend.serializers import AuditSerializer@@ -54,6 +55,14 @@ class CustomToolSerializer(IntegrityErrorMixin, AuditSerializer):}}+ def validate_tool_name(self, value: str) -> str:+ return validate_name_field(value, field_name="Tool name")++ def validate_description(self, value: str) -> str:+ if value is None:+ return value+ return validate_no_html_tags(value, field_name="Description")+def validate_summarize_llm_adapter(self, value):"""Validate that the adapter type is LLM and is accessible to the user."""if value is None:diff --git a/backend/utils/input_sanitizer.py b/backend/utils/input_sanitizer.pynew file mode 100644index 000000000..36772727e--- /dev/null+++ b/backend/utils/input_sanitizer.py@@ -0,0 +1,45 @@+import re++from rest_framework.serializers import ValidationError++# Pattern to detect HTML/script tags (closed tags and unclosed tags starting with a letter)+# The second alternative catches unclosed tags like "<script" or "<img src=x" that could+# be completed by adjacent content in non-React rendering contexts (emails, PDFs, logs)+HTML_TAG_PATTERN = re.compile(r"<[^>]*>|<[a-zA-Z/!]")+# Pattern to detect dangerous URI protocols: javascript:, vbscript:, and data: URIs.+# data: URIs are only matched when followed by a MIME type (word/word) to avoid+# false positives on ordinary English text like "Input data: JSON format".+JS_PROTOCOL_PATTERN = re.compile(+ r"(?:javascript|vbscript)\s*:|data\s*:\s*\w+/\w+", re.IGNORECASE+)+# Pattern to detect event handlers using a vetted list of DOM event names.+# This avoids false positives on benign words like "connection=", "onboarding=", etc.+_DOM_EVENTS = (+ "abort|blur|change|click|close|contextmenu|copy|cut|dblclick|drag|dragend|"+ "dragenter|dragleave|dragover|dragstart|drop|error|focus|focusin|focusout|"+ "input|invalid|keydown|keypress|keyup|load|mousedown|mouseenter|mouseleave|"+ "mousemove|mouseout|mouseover|mouseup|paste|pointerdown|pointerenter|"+ "pointerleave|pointermove|pointerout|pointerover|pointerup|reset|resize|"+ "scroll|select|submit|toggle|touchcancel|touchend|touchmove|touchstart|"+ "unload|wheel"+)+EVENT_HANDLER_PATTERN = re.compile(rf"\bon({_DOM_EVENTS})\s*=", re.IGNORECASE)+++def validate_no_html_tags(value: str, field_name: str = "This field") -> str:+ """Reject values containing HTML/script tags."""+ if HTML_TAG_PATTERN.search(value):+ raise ValidationError(f"{field_name} must not contain HTML or script tags.")+ if JS_PROTOCOL_PATTERN.search(value):+ raise ValidationError(f"{field_name} must not contain dangerous URI protocols.")+ if EVENT_HANDLER_PATTERN.search(value):+ raise ValidationError(f"{field_name} must not contain event handler attributes.")+ return value+++def validate_name_field(value: str, field_name: str = "This field") -> str:+ """Validate name/identifier fields - no HTML tags, strip whitespace."""+ value = value.strip()+ if not value:+ raise ValidationError(f"{field_name} must not be empty.")+ return validate_no_html_tags(value, field_name)diff --git a/backend/utils/tests/__init__.py b/backend/utils/tests/__init__.pynew file mode 100644index 000000000..e69de29bbdiff --git a/backend/utils/tests/test_input_sanitizer.py b/backend/utils/tests/test_input_sanitizer.pynew file mode 100644index 000000000..fe3634e72--- /dev/null+++ b/backend/utils/tests/test_input_sanitizer.py@@ -0,0 +1,150 @@+import pytest+from rest_framework.serializers import ValidationError++from utils.input_sanitizer import validate_name_field, validate_no_html_tags+++class TestValidateNoHtmlTags:+ def test_clean_input_passes(self):+ assert validate_no_html_tags("Hello World") == "Hello World"++ def test_allows_normal_special_chars(self):+ assert (+ validate_no_html_tags("My workflow (v2), test - final")+ == "My workflow (v2), test - final"+ )++ def test_allows_numbers_and_punctuation(self):+ assert validate_no_html_tags("Test 123 & more!") == "Test 123 & more!"++ def test_rejects_script_tag(self):+ with pytest.raises(ValidationError, match="must not contain HTML or script tags"):+ validate_no_html_tags("<script>alert(1)</script>")++ def test_rejects_img_tag(self):+ with pytest.raises(ValidationError, match="must not contain HTML or script tags"):+ validate_no_html_tags('<img src=x onerror=alert(1)>')++ def test_rejects_div_tag(self):+ with pytest.raises(ValidationError, match="must not contain HTML or script tags"):+ validate_no_html_tags("<div>content</div>")++ def test_rejects_self_closing_tag(self):+ with pytest.raises(ValidationError, match="must not contain HTML or script tags"):+ validate_no_html_tags("<br/>")++ def test_rejects_unclosed_script_tag(self):+ with pytest.raises(ValidationError, match="must not contain HTML or script tags"):+ validate_no_html_tags("<script")++ def test_rejects_unclosed_img_tag(self):+ with pytest.raises(ValidationError, match="must not contain HTML or script tags"):+ validate_no_html_tags("<img src=x onerror")++ def test_allows_less_than_with_number(self):+ assert validate_no_html_tags("a < 3") == "a < 3"++ def test_rejects_javascript_protocol(self):+ with pytest.raises(ValidationError, match="must not contain dangerous URI protocols"):+ validate_no_html_tags("javascript:alert(1)")++ def test_rejects_javascript_protocol_with_spaces(self):+ with pytest.raises(ValidationError, match="must not contain dangerous URI protocols"):+ validate_no_html_tags("javascript :alert(1)")++ def test_rejects_javascript_protocol_case_insensitive(self):+ with pytest.raises(ValidationError, match="must not contain dangerous URI protocols"):+ validate_no_html_tags("JAVASCRIPT:alert(1)")++ def test_rejects_data_uri(self):+ with pytest.raises(ValidationError, match="must not contain dangerous URI protocols"):+ validate_no_html_tags("data:text/html,alert(1)")++ def test_rejects_data_uri_with_mime_type(self):+ with pytest.raises(ValidationError, match="must not contain dangerous URI protocols"):+ validate_no_html_tags("data:application/javascript,code")++ def test_allows_benign_data_colon_in_text(self):+ assert validate_no_html_tags("Input data: JSON format") == "Input data: JSON format"++ def test_allows_data_colon_without_mime(self):+ assert validate_no_html_tags("my data: 100 records") == "my data: 100 records"++ def test_rejects_vbscript_protocol(self):+ with pytest.raises(ValidationError, match="must not contain dangerous URI protocols"):+ validate_no_html_tags("vbscript:MsgBox")++ def test_rejects_event_handler(self):+ with pytest.raises(+ ValidationError, match="must not contain event handler attributes"+ ):+ validate_no_html_tags("onclick=alert(1)")++ def test_rejects_event_handler_with_spaces(self):+ with pytest.raises(+ ValidationError, match="must not contain event handler attributes"+ ):+ validate_no_html_tags("onerror =alert(1)")++ def test_rejects_event_handler_case_insensitive(self):+ with pytest.raises(+ ValidationError, match="must not contain event handler attributes"+ ):+ validate_no_html_tags("ONLOAD=alert(1)")++ def test_allows_benign_connection_text(self):+ assert validate_no_html_tags("connection=ok") == "connection=ok"++ def test_allows_onboarding_text(self):+ assert validate_no_html_tags("onboarding = enabled") == "onboarding = enabled"++ def test_allows_oncall_text(self):+ assert validate_no_html_tags("oncall = primary") == "oncall = primary"++ def test_allows_condition_text(self):+ assert validate_no_html_tags("condition = good") == "condition = good"++ def test_rejects_event_handler_after_semicolon(self):+ with pytest.raises(+ ValidationError, match="must not contain event handler attributes"+ ):+ validate_no_html_tags("foo;onclick=alert(1)")++ def test_rejects_event_handler_after_ampersand(self):+ with pytest.raises(+ ValidationError, match="must not contain event handler attributes"+ ):+ validate_no_html_tags("foo&onerror=alert(1)")++ def test_custom_field_name_in_error(self):+ with pytest.raises(ValidationError, match="Workflow name"):+ validate_no_html_tags("<script>", field_name="Workflow name")+++class TestValidateNameField:+ def test_clean_name_passes(self):+ assert validate_name_field("My Workflow") == "My Workflow"++ def test_strips_whitespace(self):+ assert validate_name_field(" hello ") == "hello"++ def test_rejects_empty_after_strip(self):+ with pytest.raises(ValidationError, match="must not be empty"):+ validate_name_field(" ")++ def test_rejects_html_tags(self):+ with pytest.raises(ValidationError, match="must not contain HTML or script tags"):+ validate_name_field("<script>alert(1)</script>")++ def test_allows_hyphens_and_underscores(self):+ assert validate_name_field("my-workflow_v2") == "my-workflow_v2"++ def test_allows_periods(self):+ assert validate_name_field("config.v2") == "config.v2"++ def test_allows_parentheses_and_commas(self):+ assert validate_name_field("Test (v2), final") == "Test (v2), final"++ def test_custom_field_name_in_error(self):+ with pytest.raises(ValidationError, match="Tool name"):+ validate_name_field(" ", field_name="Tool name")diff --git a/backend/workflow_manager/workflow_v2/serializers.py b/backend/workflow_manager/workflow_v2/serializers.pyindex 386f9b81a..ed3459295 100644--- a/backend/workflow_manager/workflow_v2/serializers.py+++ b/backend/workflow_manager/workflow_v2/serializers.py@@ -14,6 +14,7 @@ from rest_framework.serializers import ()from tool_instance_v2.serializers import ToolInstanceSerializerfrom tool_instance_v2.tool_instance_helper import ToolInstanceHelper+from utils.input_sanitizer import validate_name_field, validate_no_html_tagsfrom utils.serializer.integrity_error_mixin import IntegrityErrorMixinfrom backend.constants import RequestKey@@ -46,6 +47,14 @@ class WorkflowSerializer(IntegrityErrorMixin, AuditSerializer):}}+ def validate_workflow_name(self, value: str) -> str:+ return validate_name_field(value, field_name="Workflow name")++ def validate_description(self, value: str) -> str:+ if value is None:+ return value+ return validate_no_html_tags(value, field_name="Description")+def to_representation(self, instance: Workflow) -> dict[str, str]:representation: dict[str, str] = super().to_representation(instance)representation[WorkflowKey.WF_NAME] = instance.workflow_namediff --git a/frontend/nginx.conf b/frontend/nginx.confindex f3eef1953..4e13fb122 100644--- a/frontend/nginx.conf+++ b/frontend/nginx.conf@@ -47,6 +47,14 @@ http {add_header X-Frame-Options "SAMEORIGIN" always;add_header Referrer-Policy "strict-origin-when-cross-origin" always;+ # Content Security Policy (single quoted string required by nginx add_header)+ # - 'unsafe-inline' in script-src: required for runtime-config.js injected at container start+ # - 'unsafe-inline' in style-src: required by Ant Design CSS-in-JS+ # - cdn.jsdelivr.net: Monaco Editor loads from this CDN+ # - unpkg.com: PDF.js worker+ # - PostHog, GTM, reCAPTCHA, Stripe, Product Fruits: third-party services+ add_header Content-Security-Policy "default-src 'self'; script-src 'self' 'unsafe-inline' https://cdn.jsdelivr.net https://unpkg.com https://eu.i.posthog.com https://eu-assets.i.posthog.com https://www.googletagmanager.com https://www.google.com/recaptcha/ https://www.gstatic.com/recaptcha/ https://js.stripe.com https://app.productfruits.com; style-src 'self' 'unsafe-inline'; img-src 'self' data: blob: https://eu.i.posthog.com https://eu-assets.i.posthog.com; font-src 'self' data:; connect-src 'self' wss: https://eu.i.posthog.com https://eu-assets.i.posthog.com https://www.google-analytics.com https://api.stripe.com https://app.productfruits.com; frame-src 'self' https://www.google.com/recaptcha/ https://recaptcha.google.com https://js.stripe.com https://hooks.stripe.com; worker-src 'self' blob: https://unpkg.com https://cdn.jsdelivr.net; object-src 'none'; base-uri 'self'; form-action 'self' https://checkout.stripe.com; frame-ancestors 'self'" always;+# Disable TRACE and TRACK methodsif ($request_method ~ ^(TRACE|TRACK)$) {return 405;