refactor: remove json_schema response format, simplify to json_object only

services/discord-gateway/src/modules/ai-moderation/llmClient.ts

@@ -56,15 +56,8 @@ export interface LlmCallOpts {
   temperature?: number;
   /** Top-p (defaults to 0.95). */
   top_p?: number;
-  /** Force JSON output.  When true, wraps schema in json_schema response_format. */
-  jsonResponse?:
-    | { type: "json_object" }
-    | {
-        type: "json_schema";
-        name: string;
-        schema: Record<string, unknown>;
-        strict: boolean;
-      };
+  /** Force JSON output via response_format: { type: "json_object" }. */
+  jsonResponse?: { type: "json_object" };
   /** Extra retries beyond DEFAULT_RETRIES (default 2). */
   retries?: number;
 }
@@ -91,39 +84,18 @@ export async function llmChat(
     retries = DEFAULT_RETRIES,
   } = opts;
 
-  const responseFormat:
-    | { type: "json_object" }
-    | {
-        type: "json_schema";
-        json_schema: {
-          name: string;
-          schema: Record<string, unknown>;
-          strict: boolean;
-        };
-      }
-    | undefined = jsonResponse
-    ? jsonResponse.type === "json_schema"
-      ? {
-          type: "json_schema",
-          json_schema: {
-            name: jsonResponse.name,
-            schema: jsonResponse.schema,
-            strict: jsonResponse.strict,
-          },
-        }
-      : jsonResponse
-    : undefined;
+  const params: OpenAI.Chat.Completions.ChatCompletionCreateParamsNonStreaming = {
+    model,
+    messages,
+    temperature,
+    top_p,
+    max_tokens,
+    stream: false,
+  };
 
-  const params: OpenAI.Chat.Completions.ChatCompletionCreateParamsNonStreaming =
-    {
-      model,
-      messages,
-      temperature,
-      top_p,
-      max_tokens,
-      stream: false,
-      ...(responseFormat ? { response_format: responseFormat } : {}),
-    } as OpenAI.Chat.Completions.ChatCompletionCreateParamsNonStreaming;
+  if (jsonResponse) {
+    (params as unknown as Record<string, unknown>).response_format = jsonResponse;
+  }
 
   return retryWithBackoff(
     async () => {

services/discord-gateway/src/modules/ai-moderation/llmModerationClient.ts

@@ -161,55 +161,6 @@ function deriveRecommendedAction(
   return "review";
 }
 
-/**
- * JSON Schema for OpenAI's response_format: { type: "json_schema" }.
- * This enforces the exact structure the LLM must output (R2).
- */
-const MODERATION_JSON_SCHEMA = {
-  type: "object",
-  properties: {
-    results: {
-      type: "array",
-      items: {
-        type: "object",
-        properties: {
-          message_id: { type: "string" },
-          status: { type: "string", enum: ["clean", "warn", "flagged"] },
-          flags: { type: "array", items: { type: "string" } },
-          score: { type: "number", minimum: 0, maximum: 1 },
-          analysis: { type: "string" },
-          categories: { type: "array", items: { type: "string" } },
-          severity: {
-            type: "string",
-            enum: ["none", "low", "medium", "high", "critical"],
-          },
-          confidence: { type: "number", minimum: 0, maximum: 1 },
-          recommended_action: {
-            type: "string",
-            enum: ["none", "monitor", "warn", "review", "delete", "escalate"],
-          },
-          policy_version: { type: "string" },
-          evidence: { type: "array", items: { type: "string" } },
-        },
-        required: [
-          "message_id",
-          "status",
-          "flags",
-          "score",
-          "severity",
-          "confidence",
-          "recommended_action",
-          "policy_version",
-          "evidence",
-          "analysis",
-        ],
-        additionalProperties: false,
-      },
-    },
-  },
-  required: ["results"],
-  additionalProperties: false,
-};
 
 /**
  * Helper to extract JSON from a potentially conversational or markdown-wrapped string.
@@ -736,12 +687,7 @@ async function callModerationLLM(
           const completion = await llmChat({
             messages: [{ role: "user", content }],
             max_tokens: 16384,
-            jsonResponse: {
-              type: "json_schema",
-              name: "moderation_result",
-              schema: MODERATION_JSON_SCHEMA,
-              strict: true,
-            },
+            jsonResponse: { type: "json_object" },
             retries: 0,
           });