@gitlab/gitlab-ai-provider 3.4.0 → 3.4.1

package/CHANGELOG.md

@@ -2,6 +2,11 @@
 
 All notable changes to this project will be documented in this file. See [Conventional Commits](https://conventionalcommits.org) for commit guidelines.
 
+## <small>3.4.1 (2026-02-06)</small>
+
+- Merge branch 'vg/token_refresh_fix' into 'main' ([8c1f2c4](https://gitlab.com/gitlab-org/editor-extensions/gitlab-ai-provider/commit/8c1f2c4))
+- fix: detect context overflow errors before token refresh ([ce272cd](https://gitlab.com/gitlab-org/editor-extensions/gitlab-ai-provider/commit/ce272cd))
+
 ## 3.4.0 (2026-02-02)
 
 - Merge branch 'feat/ai-gateway-headers' into 'main' ([f2d4c79](https://gitlab.com/gitlab-org/editor-extensions/gitlab-ai-provider/commit/f2d4c79))

package/dist/index.d.mts

@@ -65,6 +65,11 @@ declare class GitLabAnthropicLanguageModel implements LanguageModelV2 {
      * Check if an error is a token-related authentication error that can be retried
      */
     private isTokenError;
+    /**
+     * Check if an error is a context overflow error (prompt too long)
+     * These should NOT trigger token refresh and should be reported to the user.
+     */
+    private isContextOverflowError;
     /**
      * Convert AI SDK tools to Anthropic tool format
      */
@@ -265,6 +270,11 @@ declare class GitLabOpenAILanguageModel implements LanguageModelV2 {
     get provider(): string;
     private getOpenAIClient;
     private isTokenError;
+    /**
+     * Check if an error is a context overflow error (prompt too long)
+     * These should NOT trigger token refresh and should be reported to the user.
+     */
+    private isContextOverflowError;
     private convertTools;
     private convertToolChoice;
     private convertPrompt;
@@ -342,6 +352,11 @@ declare class GitLabError extends Error {
     isRateLimitError(): boolean;
     isForbiddenError(): boolean;
     isServerError(): boolean;
+    /**
+     * Check if this error is a context overflow error (prompt too long).
+     * These errors occur when the conversation exceeds the model's token limit.
+     */
+    isContextOverflowError(): boolean;
 }
 
 declare const gitlabOAuthTokenResponseSchema: z.ZodObject<{

package/dist/index.d.ts

@@ -65,6 +65,11 @@ declare class GitLabAnthropicLanguageModel implements LanguageModelV2 {
      * Check if an error is a token-related authentication error that can be retried
      */
     private isTokenError;
+    /**
+     * Check if an error is a context overflow error (prompt too long)
+     * These should NOT trigger token refresh and should be reported to the user.
+     */
+    private isContextOverflowError;
     /**
      * Convert AI SDK tools to Anthropic tool format
      */
@@ -265,6 +270,11 @@ declare class GitLabOpenAILanguageModel implements LanguageModelV2 {
     get provider(): string;
     private getOpenAIClient;
     private isTokenError;
+    /**
+     * Check if an error is a context overflow error (prompt too long)
+     * These should NOT trigger token refresh and should be reported to the user.
+     */
+    private isContextOverflowError;
     private convertTools;
     private convertToolChoice;
     private convertPrompt;
@@ -342,6 +352,11 @@ declare class GitLabError extends Error {
     isRateLimitError(): boolean;
     isForbiddenError(): boolean;
     isServerError(): boolean;
+    /**
+     * Check if this error is a context overflow error (prompt too long).
+     * These errors occur when the conversation exceeds the model's token limit.
+     */
+    isContextOverflowError(): boolean;
 }
 
 declare const gitlabOAuthTokenResponseSchema: z.ZodObject<{

package/dist/index.js

@@ -96,6 +96,17 @@ var GitLabError = class _GitLabError extends Error {
   isServerError() {
     return this.statusCode !== void 0 && this.statusCode >= 500;
   }
+  /**
+   * Check if this error is a context overflow error (prompt too long).
+   * These errors occur when the conversation exceeds the model's token limit.
+   */
+  isContextOverflowError() {
+    if (this.statusCode !== 400) {
+      return false;
+    }
+    const message = this.message?.toLowerCase() || "";
+    return message.includes("context overflow") || message.includes("prompt is too long") || message.includes("prompt too long") || message.includes("tokens") && message.includes("maximum");
+  }
 };
 
 // src/gitlab-direct-access.ts
@@ -266,6 +277,21 @@ var GitLabAnthropicLanguageModel = class {
     }
     return false;
   }
+  /**
+   * Check if an error is a context overflow error (prompt too long)
+   * These should NOT trigger token refresh and should be reported to the user.
+   */
+  isContextOverflowError(error) {
+    if (error instanceof import_sdk.default.APIError) {
+      if (error.status === 400) {
+        const message = error.message?.toLowerCase() || "";
+        if (message.includes("prompt is too long") || message.includes("prompt too long") || message.includes("tokens") && message.includes("maximum")) {
+          return true;
+        }
+      }
+    }
+    return false;
+  }
   /**
    * Convert AI SDK tools to Anthropic tool format
    */
@@ -443,6 +469,14 @@ var GitLabAnthropicLanguageModel = class {
         warnings: []
       };
     } catch (error) {
+      if (this.isContextOverflowError(error)) {
+        const apiError = error;
+        throw new GitLabError({
+          message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+          statusCode: 400,
+          cause: error
+        });
+      }
       if (!isRetry && this.isTokenError(error)) {
         this.directAccessClient.invalidateToken();
         return this.doGenerateWithRetry(options, true);
@@ -450,6 +484,7 @@ var GitLabAnthropicLanguageModel = class {
       if (error instanceof import_sdk.default.APIError) {
         throw new GitLabError({
           message: `Anthropic API error: ${error.message}`,
+          statusCode: error.status,
           cause: error
         });
       }
@@ -630,6 +665,19 @@ var GitLabAnthropicLanguageModel = class {
               });
             }
           }
+          if (self.isContextOverflowError(error)) {
+            const apiError = error;
+            controller.enqueue({
+              type: "error",
+              error: new GitLabError({
+                message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+                statusCode: 400,
+                cause: error
+              })
+            });
+            controller.close();
+            return;
+          }
           if (!isRetry && self.isTokenError(error)) {
             self.directAccessClient.invalidateToken();
             controller.enqueue({
@@ -647,6 +695,7 @@ var GitLabAnthropicLanguageModel = class {
               type: "error",
               error: new GitLabError({
                 message: `Anthropic API error: ${error.message}`,
+                statusCode: error.status,
                 cause: error
               })
             });
@@ -771,6 +820,21 @@ var GitLabOpenAILanguageModel = class {
     }
     return false;
   }
+  /**
+   * Check if an error is a context overflow error (prompt too long)
+   * These should NOT trigger token refresh and should be reported to the user.
+   */
+  isContextOverflowError(error) {
+    if (error instanceof import_openai.default.APIError) {
+      if (error.status === 400) {
+        const message = error.message?.toLowerCase() || "";
+        if (message.includes("prompt is too long") || message.includes("prompt too long") || message.includes("tokens") && message.includes("maximum")) {
+          return true;
+        }
+      }
+    }
+    return false;
+  }
   convertTools(tools) {
     if (!tools || tools.length === 0) {
       return void 0;
@@ -1051,6 +1115,14 @@ var GitLabOpenAILanguageModel = class {
         warnings: []
       };
     } catch (error) {
+      if (this.isContextOverflowError(error)) {
+        const apiError = error;
+        throw new GitLabError({
+          message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+          statusCode: 400,
+          cause: error
+        });
+      }
       if (!isRetry && this.isTokenError(error)) {
         this.directAccessClient.invalidateToken();
         return this.doGenerateWithChatApi(options, true);
@@ -1058,6 +1130,7 @@ var GitLabOpenAILanguageModel = class {
       if (error instanceof import_openai.default.APIError) {
         throw new GitLabError({
           message: `OpenAI API error: ${error.message}`,
+          statusCode: error.status,
           cause: error
         });
       }
@@ -1113,6 +1186,14 @@ var GitLabOpenAILanguageModel = class {
         warnings: []
       };
     } catch (error) {
+      if (this.isContextOverflowError(error)) {
+        const apiError = error;
+        throw new GitLabError({
+          message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+          statusCode: 400,
+          cause: error
+        });
+      }
       if (!isRetry && this.isTokenError(error)) {
         this.directAccessClient.invalidateToken();
         return this.doGenerateWithResponsesApi(options, true);
@@ -1120,6 +1201,7 @@ var GitLabOpenAILanguageModel = class {
       if (error instanceof import_openai.default.APIError) {
         throw new GitLabError({
           message: `OpenAI API error: ${error.message}`,
+          statusCode: error.status,
           cause: error
         });
       }
@@ -1238,6 +1320,19 @@ var GitLabOpenAILanguageModel = class {
           controller.enqueue({ type: "finish", finishReason, usage });
           controller.close();
         } catch (error) {
+          if (self.isContextOverflowError(error)) {
+            const apiError = error;
+            controller.enqueue({
+              type: "error",
+              error: new GitLabError({
+                message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+                statusCode: 400,
+                cause: error
+              })
+            });
+            controller.close();
+            return;
+          }
           if (!isRetry && self.isTokenError(error)) {
             self.directAccessClient.invalidateToken();
             controller.enqueue({
@@ -1252,6 +1347,7 @@ var GitLabOpenAILanguageModel = class {
               type: "error",
               error: new GitLabError({
                 message: `OpenAI API error: ${error.message}`,
+                statusCode: error.status,
                 cause: error
               })
             });
@@ -1378,6 +1474,19 @@ var GitLabOpenAILanguageModel = class {
           controller.enqueue({ type: "finish", finishReason, usage });
           controller.close();
         } catch (error) {
+          if (self.isContextOverflowError(error)) {
+            const apiError = error;
+            controller.enqueue({
+              type: "error",
+              error: new GitLabError({
+                message: `Context overflow: ${apiError.message}. Please start a new session or use /compact to reduce context.`,
+                statusCode: 400,
+                cause: error
+              })
+            });
+            controller.close();
+            return;
+          }
           if (!isRetry && self.isTokenError(error)) {
             self.directAccessClient.invalidateToken();
             controller.enqueue({
@@ -1392,6 +1501,7 @@ var GitLabOpenAILanguageModel = class {
               type: "error",
               error: new GitLabError({
                 message: `OpenAI API error: ${error.message}`,
+                statusCode: error.status,
                 cause: error
               })
             });
@@ -1569,7 +1679,7 @@ var GitLabOAuthManager = class {
 };
 
 // src/version.ts
-var VERSION = true ? "3.3.1" : "0.0.0-dev";
+var VERSION = true ? "3.4.0" : "0.0.0-dev";
 
 // src/gitlab-provider.ts
 var fs = __toESM(require("fs"));