Merge pull request #1483 from qualifire-dev/feature/qualifire-update-apis

Qualifire | Update Guardrails

Author: VisargD

plugins/index.ts

@@ -14,15 +14,11 @@ import { handler as defaultalllowercase } from './default/alllowercase';
 import { handler as defaultendsWith } from './default/endsWith';
 import { handler as defaultmodelWhitelist } from './default/modelWhitelist';
 import { handler as defaultnotNull } from './default/notNull';
-import { handler as qualifireDangerousContent } from './qualifire/dangerousContent';
+import { handler as qualifireContentModeration } from './qualifire/contentModeration';
 import { handler as qualifireGrounding } from './qualifire/grounding';
-import { handler as qualifireHarassment } from './qualifire/harassment';
-import { handler as qualifireInstructionFollowing } from './qualifire/instructionFollowing';
 import { handler as qualifirePolicy } from './qualifire/policy';
-import { handler as qualifireSexualContent } from './qualifire/sexualContent';
 import { handler as qualifireToolUseQuality } from './qualifire/toolUseQuality';
 import { handler as qualifireHallucinations } from './qualifire/hallucinations';
-import { handler as qualifireHateSpeech } from './qualifire/hateSpeech';
 import { handler as qualifirePii } from './qualifire/pii';
 import { handler as qualifirePromptInjections } from './qualifire/promptInjections';
 import { handler as defaultaddPrefix } from './default/addPrefix';
@@ -97,15 +93,11 @@ export const plugins = {
     notNull: defaultnotNull,
   },
   qualifire: {
-    dangerousContent: qualifireDangerousContent,
+    contentModeration: qualifireContentModeration,
     grounding: qualifireGrounding,
-    harassment: qualifireHarassment,
-    instructionFollowing: qualifireInstructionFollowing,
     policy: qualifirePolicy,
-    sexualContent: qualifireSexualContent,
     toolUseQuality: qualifireToolUseQuality,
     hallucinations: qualifireHallucinations,
-    hateSpeech: qualifireHateSpeech,
     pii: qualifirePii,
     promptInjections: qualifirePromptInjections,
   },

plugins/qualifire/dangerousContent.ts plugins/qualifire/contentModeration.tsplugins/qualifire/dangerousContent.ts renamed to plugins/qualifire/contentModeration.ts

@@ -13,7 +13,7 @@ export const handler: PluginHandler = async (
 ) => {
   const evaluationBody: any = {
     input: context.request.text,
-    dangerous_content_check: true,
+    content_moderation_check: true,
   };
 
   if (eventType === 'afterRequestHook') {

plugins/qualifire/globals.ts

@@ -38,7 +38,7 @@ export const postQualifire = async (
 
   const result = await post(BASE_URL, body, options, timeout_millis || 60000);
   const error = result?.error || null;
-  const verdict = result?.status === 'success';
+  const verdict = result?.status !== 'failed';
   const data = result?.evaluationResults;
 
   return { error, verdict, data };

plugins/qualifire/grounding.ts

@@ -22,10 +22,13 @@ export const handler: PluginHandler = async (
     };
   }
 
+  const mode = parameters?.mode || 'balanced';
+
   const evaluationBody: any = {
     input: context.request.text,
     output: context.response.text,
     grounding_check: true,
+    grounding_mode: mode,
   };
 
   try {

plugins/qualifire/hallucinations.ts

@@ -22,10 +22,13 @@ export const handler: PluginHandler = async (
     };
   }
 
+  const mode = parameters?.mode || 'balanced';
+
   const evaluationBody: any = {
     input: context.request.text,
     output: context.response.text,
     hallucinations_check: true,
+    hallucinations_mode: mode,
   };
 
   try {

plugins/qualifire/harassment.ts

@@ -15,66 +15,14 @@
   },
   "functions": [
     {
-      "name": "Hate Speech Check",
-      "id": "hateSpeech",
+      "name": "Content Moderation Check",
+      "id": "contentModeration",
       "supportedHooks": ["beforeRequestHook", "afterRequestHook"],
       "type": "guardrail",
       "description": [
         {
           "type": "subHeading",
-          "text": "Checks for hate speech in the user input or model output."
-        }
-      ],
-      "parameters": {}
-    },
-    {
-      "name": "Dangerous Content Check",
-      "id": "dangerousContent",
-      "supportedHooks": ["beforeRequestHook", "afterRequestHook"],
-      "type": "guardrail",
-      "description": [
-        {
-          "type": "subHeading",
-          "text": "Checks for dangerous content in the user input or model output."
-        }
-      ],
-      "parameters": {}
-    },
-    {
-      "name": "Sexual Content Check",
-      "id": "sexualContent",
-      "supportedHooks": ["beforeRequestHook", "afterRequestHook"],
-      "type": "guardrail",
-      "description": [
-        {
-          "type": "subHeading",
-          "text": "Checks for sexual content in the user input or model output."
-        }
-      ],
-      "parameters": {}
-    },
-    {
-      "name": "Harassment Check",
-      "id": "harassment",
-      "supportedHooks": ["beforeRequestHook", "afterRequestHook"],
-      "type": "guardrail",
-      "description": [
-        {
-          "type": "subHeading",
-          "text": "Checks for harassment in the user input or model output."
-        }
-      ],
-      "parameters": {}
-    },
-    {
-      "name": "Instruction Following Check",
-      "id": "instructionFollowing",
-      "supportedHooks": ["afterRequestHook"],
-      "type": "guardrail",
-      "description": [
-        {
-          "type": "subHeading",
-          "text": "Checks that the model followed the instructions provided in the prompt."
+          "text": "Checks for dangerous content, sexual content, harassment, and dangerous content in the user input or model output."
         }
       ],
       "parameters": {}
@@ -90,7 +38,18 @@
           "text": "Checks that the model did not hallucinate."
         }
       ],
-      "parameters": {}
+      "parameters": {
+        "type": "object",
+        "properties": {
+          "mode": {
+            "type": "string",
+            "label": "Mode",
+            "description": "The mode to use for the check",
+            "enum": ["speed", "balanced", "quality"],
+            "default": "balanced"
+          }
+        }
+      }
     },
     {
       "name": "PII Check",
@@ -129,7 +88,18 @@
           "text": "Checks that the model is grounded in the context provided."
         }
       ],
-      "parameters": {}
+      "parameters": {
+        "type": "object",
+        "properties": {
+          "mode": {
+            "type": "string",
+            "label": "Mode",
+            "description": "The mode to use for the check",
+            "enum": ["speed", "balanced", "quality"],
+            "default": "balanced"
+          }
+        }
+      }
     },
     {
       "name": "Tool Use Quality Check",
@@ -142,7 +112,18 @@
           "text": "Checks the model's tool use quality. Including correct tool selection, correct tool parameters and values."
         }
       ],
-      "parameters": {}
+      "parameters": {
+        "type": "object",
+        "properties": {
+          "mode": {
+            "type": "string",
+            "label": "Mode",
+            "description": "The mode to use for the check",
+            "enum": ["speed", "balanced", "quality"],
+            "default": "balanced"
+          }
+        }
+      }
     },
     {
       "name": "Policy Violations Check",
@@ -170,6 +151,20 @@
                 }
               ]
             }
+          },
+          "mode": {
+            "type": "string",
+            "label": "Mode",
+            "description": "The mode to use for the check",
+            "enum": ["speed", "balanced", "quality"],
+            "default": "balanced"
+          },
+          "policy_target": {
+            "type": "string",
+            "label": "Policy Target",
+            "description": "Where to apply the policy check",
+            "enum": ["input", "output", "both"],
+            "default": "both"
           }
         },
         "required": ["policies"]

plugins/qualifire/hateSpeech.ts

@@ -21,12 +21,24 @@ export const handler: PluginHandler = async (
     };
   }
 
+  const mode = parameters?.mode || 'balanced';
+  const policyTarget = parameters?.policy_target || 'both';
+
   const evaluationBody: any = {
-    input: context.request.text,
     assertions: parameters?.policies,
+    assertions_mode: mode,
   };
 
-  if (eventType === 'afterRequestHook') {
+  // Add input based on policy_target
+  if (policyTarget === 'input' || policyTarget === 'both') {
+    evaluationBody.input = context.request.text;
+  }
+
+  // Add output based on policy_target and hook type
+  if (
+    eventType === 'afterRequestHook' &&
+    (policyTarget === 'output' || policyTarget === 'both')
+  ) {
     evaluationBody.output = context.response.text;
   }