Add Base Prompt Rule functionality to Guardrail Policies

apps/dashboard/src/main/java/com/akto/action/GuardrailPoliciesAction.java

@@ -1,7 +1,9 @@
 package com.akto.action;
 
+import com.akto.dao.ApiCollectionsDao;
 import com.akto.dao.GuardrailPoliciesDao;
 import com.akto.dao.context.Context;
+import com.akto.dto.ApiCollection;
 import com.akto.dto.GuardrailPolicies;
 import com.akto.dto.User;
 import com.akto.log.LoggerMaker;
@@ -10,6 +12,7 @@
 import com.mongodb.client.model.Filters;
 import com.mongodb.client.model.UpdateOptions;
 import com.mongodb.client.model.Updates;
+import org.apache.commons.lang3.StringUtils;
 
 import lombok.Getter;
 import lombok.Setter;
@@ -20,6 +23,7 @@
 import java.util.List;
 import java.util.Map;
 
+
 public class GuardrailPoliciesAction extends UserAction {
     private static final LoggerMaker loggerMaker = new LoggerMaker(GuardrailPoliciesAction.class, LogDb.DASHBOARD);
 
@@ -43,6 +47,11 @@ public String fetchGuardrailPolicies() {
             this.guardrailPolicies  = GuardrailPoliciesDao.instance.findAllSortedByCreatedTimestamp(0, 20);
             this.total = GuardrailPoliciesDao.instance.getTotalCount();
 
+            // Populate basePrompt for policies with autoDetect enabled
+            for (GuardrailPolicies policy : guardrailPolicies) {
+                populateBasePromptIfNeeded(policy);
+            }
+
             loggerMaker.info("Fetched " + guardrailPolicies.size() + " guardrail policies out of " + total + " total");
 
             return SUCCESS.toUpperCase();
@@ -52,6 +61,53 @@ public String fetchGuardrailPolicies() {
         }
     }
 
+    /**
+     * Populates basePrompt in basePromptRule if:
+     * 1. basePromptRule exists and is enabled
+     * 2. autoDetect is true
+     * 3. basePrompt is not already set (or is empty)
+     * 4. There are selected agent servers
+     * 
+     * Fetches detectedBasePrompt from the first selected agent collection
+     */
+    private void populateBasePromptIfNeeded(GuardrailPolicies policy) {
+        try {
+            GuardrailPolicies.BasePromptRule basePromptRule = policy.getBasePromptRule();
+            if (basePromptRule == null || !basePromptRule.isEnabled() || !basePromptRule.isAutoDetect()) {
+                return;
+            }
+
+            // If basePrompt is already set, use it
+            if (StringUtils.isNotBlank(basePromptRule.getBasePrompt())) {
+                return;
+            }
+
+            // Get selected agent servers (prefer V2 format, fallback to old format)
+            List<GuardrailPolicies.SelectedServer> agentServers = policy.getEffectiveSelectedAgentServers();
+            if (agentServers == null || agentServers.isEmpty()) {
+                return;
+            }
+
+            // Try to fetch detected base prompt from the first selected agent collection
+            try {
+                int firstAgentCollectionId = Integer.parseInt(agentServers.get(0).getId());
+                ApiCollection agentCollection = ApiCollectionsDao.instance.getMeta(firstAgentCollectionId);
+
+                if (agentCollection != null && StringUtils.isNotBlank(agentCollection.getDetectedBasePrompt())) {
+                    basePromptRule.setBasePrompt(agentCollection.getDetectedBasePrompt());
+                    loggerMaker.debug("Populated basePrompt from collection " + firstAgentCollectionId + 
+                        " for policy: " + policy.getName());
+                }
+            } catch (NumberFormatException e) {
+                loggerMaker.debug("Invalid agent collection ID format: " + agentServers.get(0).getId());
+            } catch (Exception e) {
+                loggerMaker.debug("Error fetching detected base prompt for policy " + policy.getName() + ": " + e.getMessage());
+            }
+        } catch (Exception e) {
+            loggerMaker.debug("Error populating base prompt: " + e.getMessage());
+        }
+    }
+
     public String createGuardrailPolicy() {
         try {
             User user = getSUser();
@@ -96,6 +152,7 @@ public String createGuardrailPolicy() {
             updates.add(Updates.set("regexPatternsV2", policy.getRegexPatternsV2()));
             updates.add(Updates.set("contentFiltering", policy.getContentFiltering()));
             updates.add(Updates.set("llmRule", policy.getLlmRule()));
+            updates.add(Updates.set("basePromptRule", policy.getBasePromptRule()));
             updates.add(Updates.set("selectedMcpServers", policy.getSelectedMcpServers()));
             updates.add(Updates.set("selectedAgentServers", policy.getSelectedAgentServers()));
             updates.add(Updates.set("selectedMcpServersV2", policy.getSelectedMcpServersV2()));

apps/dashboard/web/polaris_web/web/src/apps/dashboard/pages/guardrails/GuardrailPolicies.jsx

@@ -430,6 +430,8 @@ function GuardrailPolicies() {
                 contentFiltering: guardrailData.contentFilters || {},
                 // Add LLM policy if present
                 ...(guardrailData.llmRule ? { llmRule: guardrailData.llmRule } : {}),
+                // Add Base Prompt Rule if present
+                ...(guardrailData.basePromptRule ? { basePromptRule: guardrailData.basePromptRule } : {}),
                 applyOnResponse: guardrailData.applyOnResponse || false,
                 applyOnRequest: guardrailData.applyOnRequest || false,
                 url: guardrailData.url || '',

apps/dashboard/web/polaris_web/web/src/apps/dashboard/pages/guardrails/components/CreateGuardrailModal.jsx

@@ -76,6 +76,12 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
     const [enableLlmRule, setEnableLlmRule] = useState(false);
     const [llmConfidenceScore, setLlmConfidenceScore] = useState(0.5);
 
+    // Step 6.5: Base Prompt Rule
+    const [enableBasePromptRule, setEnableBasePromptRule] = useState(false);
+    const [basePrompt, setBasePrompt] = useState("");
+    const [basePromptAutoDetect, setBasePromptAutoDetect] = useState(true);
+    const [basePromptConfidenceScore, setBasePromptConfidenceScore] = useState(0.5);
+
     // Step 7: URL and Confidence Score
     const [url, setUrl] = useState("");
     const [confidenceScore, setConfidenceScore] = useState(25); // Start with 25 (first checkpoint)
@@ -162,12 +168,18 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
         },
         {
             number: 7,
+            title: "Base Prompt Rule",
+            optional: true,
+            summary: enableBasePromptRule ? `Enabled${basePromptAutoDetect ? ' (Auto-detect)' : ''}${basePrompt ? ` - ${basePrompt.substring(0, 30)}${basePrompt.length > 30 ? '...' : ''}` : ''}, Confidence: ${basePromptConfidenceScore.toFixed(2)}` : null
+        },
+        {
+            number: 8,
             title: "URL and Confidence Score",
             optional: true,
             summary: url ? `URL: ${url.substring(0, 30)}${url.length > 30 ? '...' : ''}, Confidence: ${confidenceScore}` : null
         },
         {
-            number: 8,
+            number: 9,
             title: "Server and application settings",
             optional: false,
             summary: (selectedMcpServers.length > 0 || selectedAgentServers.length > 0)
@@ -289,6 +301,10 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
         setLlmRule("");
         setEnableLlmRule(false);
         setLlmConfidenceScore(0.5);
+        setEnableBasePromptRule(false);
+        setBasePrompt("");
+        setBasePromptAutoDetect(true);
+        setBasePromptConfidenceScore(0.5);
         setUrl("");
         setConfidenceScore(25);
         setUrlError("");
@@ -359,6 +375,19 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
             setLlmConfidenceScore(0.5);
         }
 
+        // Base Prompt Rule
+        if (policy.basePromptRule) {
+            setEnableBasePromptRule(policy.basePromptRule.enabled || false);
+            setBasePrompt(policy.basePromptRule.basePrompt || "");
+            setBasePromptAutoDetect(policy.basePromptRule.autoDetect !== undefined ? policy.basePromptRule.autoDetect : true);
+            setBasePromptConfidenceScore(policy.basePromptRule.confidenceScore !== undefined ? policy.basePromptRule.confidenceScore : 0.5);
+        } else {
+            setEnableBasePromptRule(false);
+            setBasePrompt("");
+            setBasePromptAutoDetect(true);
+            setBasePromptConfidenceScore(0.5);
+        }
+
         // URL and Confidence Score
         setUrl(policy.url || "");
         // Map existing confidence score to nearest checkpoint
@@ -406,7 +435,7 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
     };
 
     const handleSkipToServers = () => {
-        setCurrentStep(8);
+        setCurrentStep(9);
     };
 
     const handleSave = async () => {
@@ -465,6 +494,14 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
                         confidenceScore: llmConfidenceScore
                     }
                 } : {}),
+                ...(enableBasePromptRule ? {
+                    basePromptRule: {
+                        enabled: true,
+                        basePrompt: basePromptAutoDetect ? null : basePrompt.trim(), // Only save manual prompt, auto-detect will fetch on read
+                        autoDetect: basePromptAutoDetect,
+                        confidenceScore: basePromptConfidenceScore
+                    }
+                } : {}),
                 url: url || null,
                 confidenceScore: confidenceScore,
                 selectedMcpServers: selectedMcpServers, // Old format (just IDs)
@@ -958,6 +995,62 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
     );
 
     const renderStep7 = () => (
+        <LegacyCard sectioned>
+            <VerticalStack gap="4">
+                <Text variant="headingMd">Base Prompt Rule</Text>
+                <Text variant="bodyMd" tone="subdued">
+                    Configure a base prompt rule to check the intent of user input in agent prompts with placeholders like {`{var}`} or {`{}`}.
+                </Text>
+
+                <Checkbox
+                    label="Enable base prompt rule"
+                    checked={enableBasePromptRule}
+                    onChange={setEnableBasePromptRule}
+                    helpText="When enabled, the guardrail will analyze user inputs that fill placeholders in the base prompt."
+                />
+
+                {enableBasePromptRule && (
+                    <>
+                        <Checkbox
+                            label="Auto-detect base prompt from traffic"
+                            checked={basePromptAutoDetect}
+                            onChange={setBasePromptAutoDetect}
+                            helpText="Automatically detect the base prompt pattern from agent traffic. If disabled, you must provide the base prompt manually."
+                        />
+
+                        {!basePromptAutoDetect && (
+                            <TextField
+                                label="Base Prompt Template"
+                                value={basePrompt}
+                                onChange={setBasePrompt}
+                                multiline={5}
+                                placeholder="You are a helpful assistant. Answer the following question: {}"
+                                helpText="Provide the base prompt template with placeholders using {} or {var_name} syntax."
+                            />
+                        )}
+
+                        <Box>
+                            <Text variant="bodyMd" fontWeight="medium">Confidence Score: {basePromptConfidenceScore.toFixed(2)}</Text>
+                            <Box paddingBlockStart="2">
+                                <RangeSlider
+                                    label=""
+                                    value={basePromptConfidenceScore}
+                                    min={0}
+                                    max={1}
+                                    step={0.01}
+                                    output
+                                    onChange={setBasePromptConfidenceScore}
+                                    helpText="Set the confidence threshold (0-1). Higher values require more confidence to block content."
+                                />
+                            </Box>
+                        </Box>
+                    </>
+                )}
+            </VerticalStack>
+        </LegacyCard>
+    );
+
+    const renderStep8 = () => (
         <LegacyCard sectioned>
             <VerticalStack gap="4">
                 <Text variant="headingMd">URL and Confidence Score</Text>
@@ -1001,7 +1094,7 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
         </LegacyCard>
     );
 
-    const renderStep8 = () => (
+    const renderStep9 = () => (
         <LegacyCard sectioned>
             <VerticalStack gap="4">
                 <Text variant="headingMd">Server and application settings</Text>
@@ -1069,6 +1162,7 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
             case 6: return renderStep6();
             case 7: return renderStep7();
             case 8: return renderStep8();
+            case 9: return renderStep9();
             default: return renderStep1();
         }
     };
@@ -1083,7 +1177,7 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
             });
         }
 
-        if (currentStep > 1 && currentStep < 8) {
+        if (currentStep > 1 && currentStep < 9) {
             actions.push({
                 content: "Skip to Server settings",
                 onAction: handleSkipToServers
@@ -1094,14 +1188,14 @@ const CreateGuardrailModal = ({ isOpen, onClose, onSave, editingPolicy = null, i
     };
 
     const getPrimaryAction = () => {
-        if (currentStep === 8) {
+        if (currentStep === 9) {
             return {
                 content: isEditMode ? "Update Guardrail" : "Create Guardrail",
                 onAction: handleSave,
                 loading: loading,
                 disabled: !name.trim() || !blockedMessage.trim() || urlError
             };
-        } else if (currentStep < 8) {
+        } else if (currentStep < 9) {
             return {
                 content: "Next",
                 onAction: handleNext,

libs/dao/src/main/java/com/akto/dto/ApiCollection.java

@@ -72,6 +72,9 @@ public class ApiCollection {
     String registryStatus;
     public static final String REGISTRY_STATUS = "registryStatus";
 
+    String detectedBasePrompt;
+    public static final String DETECTED_BASE_PROMPT = "detectedBasePrompt";
+
     private static final List<String> ENV_KEYWORDS_WITH_DOT = Arrays.asList(
         "staging", "preprod", "qa", "demo", "dev", "test", "svc", 
         "localhost", "local", "intranet", "lan", "example", "invalid", 
@@ -467,4 +470,12 @@ public String getRegistryStatus() {
     public void setRegistryStatus(String registryStatus) {
         this.registryStatus = registryStatus;
     }
+
+    public String getDetectedBasePrompt() {
+        return detectedBasePrompt;
+    }
+
+    public void setDetectedBasePrompt(String detectedBasePrompt) {
+        this.detectedBasePrompt = detectedBasePrompt;
+    }
 }

libs/dao/src/main/java/com/akto/dto/GuardrailPolicies.java

@@ -50,6 +50,9 @@ public class GuardrailPolicies {
 
     private LLMRule llmRule;
 
+    // Step 6.5: Base Prompt Rule - for checking intent of user input in agent base prompts with placeholders
+    private BasePromptRule basePromptRule;
+
     // Step 7: Server and application settings (old format - backward compatibility)
     private List<String> selectedMcpServers;
     private List<String> selectedAgentServers;
@@ -118,7 +121,7 @@ public GuardrailPolicies(String name, String description, String blockedMessage,
                            int updatedTimestamp, String createdBy, String updatedBy, String selectedCollection,
                            String selectedModel, List<DeniedTopic> deniedTopics, List<PiiType> piiTypes,
                            List<String> regexPatterns, List<RegexPattern> regexPatternsV2, Map<String, Object> contentFiltering,
-                           LLMRule llmRule,
+                           LLMRule llmRule, BasePromptRule basePromptRule,
                            List<String> selectedMcpServers, List<String> selectedAgentServers,
                            List<SelectedServer> selectedMcpServersV2, List<SelectedServer> selectedAgentServersV2,
                            boolean applyOnResponse, boolean applyOnRequest, String url, double confidenceScore, boolean active) {
@@ -138,6 +141,7 @@ public GuardrailPolicies(String name, String description, String blockedMessage,
         this.regexPatternsV2 = regexPatternsV2;
         this.contentFiltering = contentFiltering;
         this.llmRule = llmRule;
+        this.basePromptRule = basePromptRule;
         this.selectedMcpServers = selectedMcpServers;
         this.selectedAgentServers = selectedAgentServers;
         this.selectedMcpServersV2 = selectedMcpServersV2;
@@ -217,4 +221,21 @@ public LLMRule(boolean enabled, String userPrompt, double confidenceScore) {
             this.confidenceScore = confidenceScore;
         }
     }
+
+    @Getter
+    @Setter
+    @NoArgsConstructor
+    public static class BasePromptRule {
+        private boolean enabled;
+        private String basePrompt; // Base prompt with placeholders like {var} or {}
+        private boolean autoDetect; // Whether to auto-detect base_prompt from traffic
+        private double confidenceScore;
+
+        public BasePromptRule(boolean enabled, String basePrompt, boolean autoDetect, double confidenceScore) {
+            this.enabled = enabled;
+            this.basePrompt = basePrompt;
+            this.autoDetect = autoDetect;
+            this.confidenceScore = confidenceScore;
+        }
+    }
 }