diff --git a/CHANGELOG.md b/CHANGELOG.md
index 08698a0..df48b94 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,12 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [Unreleased]
+
+### Fixed
+
+- **Silent permission reply failures** ([#35](https://github.com/ben-vargas/ai-sdk-provider-opencode-sdk/pull/35)) - `replyToPendingApprovals` never inspected the resolved value of `permission.reply`. Managed clients use `responseStyle: "fields"` without `throwOnError`, so API-level failures resolve as `{ error }` instead of throwing — a failed reply was silently recorded as replied and never retried, leaving OpenCode waiting on the permission request. The result is now checked via `extractSdkResult` (matching the question-reply handling): on error, a warning is logged and surfaced in the response `warnings`, and the approval id is not recorded as replied so the next turn retries it.
+
 ## [3.0.6] - 2026-06-11
 
 ### Fixed
diff --git a/src/opencode-language-model.test.ts b/src/opencode-language-model.test.ts
index e7d0282..2f709f8 100644
--- a/src/opencode-language-model.test.ts
+++ b/src/opencode-language-model.test.ts
@@ -634,6 +634,55 @@ describe("opencode-language-model", () => {
       expect(mockClient.permission.reply).toHaveBeenCalledTimes(1);
     });
 
+    it("should warn and not record the approval as replied when permission.reply returns an API error", async () => {
+      mockClient.permission.reply.mockResolvedValueOnce({
+        error: { message: "unknown permission request" },
+      });
+      const logger = { warn: vi.fn(), error: vi.fn() };
+      model = new OpencodeLanguageModel({
+        modelId: "anthropic/claude-3-5-sonnet-20241022",
+        settings: { logger },
+        clientManager: mockClientManager as unknown as OpencodeClientManager,
+      });
+      const promptWithApproval: LanguageModelV3Prompt = [
+        {
+          role: "user",
+          content: [{ type: "text", text: "Continue after approval." }],
+        },
+        {
+          role: "tool",
+          content: [
+            {
+              type: "tool-approval-response",
+              approvalId: "approval-1",
+              approved: true,
+            },
+          ],
+        },
+      ];
+
+      const result = await model.doGenerate({
+        prompt: promptWithApproval,
+      });
+
+      const expectedWarning = expect.stringContaining(
+        "Failed to apply tool approval response for approval-1",
+      );
+      expect(logger.warn).toHaveBeenCalledWith(expectedWarning);
+      expect(result.warnings).toContainEqual({
+        type: "other",
+        message: expectedWarning,
+      });
+
+      // The failed reply must not be recorded as replied, so the next turn
+      // retries it.
+      await model.doGenerate({
+        prompt: promptWithApproval,
+      });
+
+      expect(mockClient.permission.reply).toHaveBeenCalledTimes(2);
+    });
+
     it("should not emit duplicate document sources for local files with source metadata", async () => {
       mockClient.session.prompt.mockResolvedValueOnce({
         data: {
diff --git a/src/opencode-language-model.ts b/src/opencode-language-model.ts
index 5204fc0..91aef88 100644
--- a/src/opencode-language-model.ts
+++ b/src/opencode-language-model.ts
@@ -750,12 +750,26 @@ export class OpencodeLanguageModel implements LanguageModelV3 {
 
     for (const response of pendingResponses) {
       try {
-        await permissionApi.reply({
+        const result = await permissionApi.reply({
           requestID: response.approvalId,
           reply: response.approved ? "once" : "reject",
           ...(response.reason ? { message: response.reason } : {}),
           ...(directory ? { directory } : {}),
         });
+
+        // Fields-style clients report API failures via the result rather
+        // than throwing, so a missing check would record the approval as
+        // replied while OpenCode keeps waiting on the permission request.
+        const { error: resultError } = extractSdkResult(result);
+        if (resultError) {
+          const warning =
+            `Failed to apply tool approval response for ${response.approvalId}: ` +
+            `${extractErrorMessage(resultError)}`;
+          this.logger.warn(warning);
+          warnings.push(warning);
+          continue;
+        }
+
         repliedApprovalIds.add(response.approvalId);
       } catch (error) {
         const warning =