diff --git a/src/adapters/openai-responses.ts b/src/adapters/openai-responses.ts
index d9ec1fb01a..a0aed2c608 100644
--- a/src/adapters/openai-responses.ts
+++ b/src/adapters/openai-responses.ts
@@ -906,6 +906,61 @@ function toolOutputText(output: unknown): string {
   }).filter(Boolean).join("\n");
 }
 
+/** True when an output can be losslessly represented as user-message content. */
+function isRepairableToolOutput(output: unknown): output is string | Record<string, unknown>[] {
+  if (typeof output === "string") return true;
+  if (!Array.isArray(output)) return false;
+  return output.every(part => {
+    if (!isPlainObject(part)) return false;
+    if (typeof part.type !== "string") return false;
+    if (["output_text", "text", "input_text"].includes(part.type)) {
+      return typeof part.text === "string";
+    }
+    if (part.type === "refusal") return typeof part.refusal === "string";
+    if (part.type === "encrypted_content") return typeof part.encrypted_content === "string";
+    if (part.type !== "input_image") return false;
+    const imageUrl = part.image_url;
+    const fileId = part.file_id;
+    const imageUrlIsString = typeof imageUrl === "string";
+    const fileIdIsString = typeof fileId === "string";
+    const hasUsableSource = (imageUrlIsString && imageUrl.length > 0)
+      || (fileIdIsString && fileId.length > 0);
+    const validSource = hasUsableSource
+      && (part.image_url === undefined || imageUrlIsString)
+      && (part.file_id === undefined || fileIdIsString);
+    const validDetail = part.detail === undefined
+      || (typeof part.detail === "string"
+        && ["auto", "low", "high", "original"].includes(part.detail));
+    return validSource && validDetail;
+  });
+}
+
+/** Convert orphaned tool output to user-message content without discarding valid images. */
+function orphanedToolOutputContent(output: unknown, callId = ""): Record<string, unknown>[] {
+  const marker = `[tool output for ${callId || "unknown call"}]`;
+  if (typeof output !== "string" && !Array.isArray(output)) {
+    return [{ type: "input_text", text: marker }];
+  }
+  if (!Array.isArray(output)) {
+    return [{ type: "input_text", text: `${marker}\n${toolOutputText(output)}` }];
+  }
+
+  const content: Record<string, unknown>[] = [{ type: "input_text", text: marker }];
+  for (const part of output) {
+    if (!isPlainObject(part)) continue;
+    if (part.type === "input_image") {
+      content.push(part);
+    } else if (part.type === "encrypted_content" && typeof part.encrypted_content === "string") {
+      content.push({ type: "input_text", text: "[encrypted content omitted]" });
+    } else if (typeof part.text === "string") {
+      content.push({ type: "input_text", text: part.text });
+    } else if (part.type === "refusal" && typeof part.refusal === "string") {
+      content.push({ type: "input_text", text: `[refusal] ${part.refusal}` });
+    }
+  }
+  return content;
+}
+
 /** True when a Responses tool output item is present but carries no usable content. */
 function isToolOutputEmpty(output: unknown): boolean {
   if (typeof output === "string") return output.trim() === "";
@@ -940,6 +995,32 @@ function annotateEmptyResponsesToolOutputs(body: unknown, enabled: boolean): unk
   return changed ? { ...body, input } : body;
 }
 
+/**
+ * Preserve the text of structurally invalid tool-output items before they reach a strict
+ * Responses parser. Stateful destinations may legitimately receive an output whose matching
+ * call lives behind `previous_response_id`, so ordinary orphan repair cannot run universally.
+ * A missing or empty `call_id`, however, cannot identify stored state on any destination.
+ */
+function repairUnidentifiedToolOutputItems(body: unknown): unknown {
+  if (!isPlainObject(body) || !Array.isArray(body.input)) return body;
+  let changed = false;
+  const input = body.input.map(item => {
+    if (!isPlainObject(item)
+      || (item.type !== "function_call_output" && item.type !== "custom_tool_call_output")
+      || (typeof item.call_id === "string" && item.call_id.length > 0)) {
+      return item;
+    }
+    if (!isRepairableToolOutput(item.output)) return item;
+    changed = true;
+    return {
+      type: "message",
+      role: "user",
+      content: orphanedToolOutputContent(item.output),
+    };
+  });
+  return changed ? { ...body, input } : body;
+}
+
 /**
  * Repair a forward-mode input array whose continuation context was lost. When the replay
  * expansion misses (proxy restart, unrecorded prior turn), previous_response_id is stripped
@@ -1060,12 +1141,17 @@ function repairOrphanedInputItems(body: unknown, dropReasoning: boolean, synthes
       flushPendingSyntheticOutputs();
       const callId = typeof item.call_id === "string" ? item.call_id : "";
       const paired = isFnOutput ? functionCallIds.has(callId) : customCallIds.has(callId);
-      if (!paired) {
+      const usableOutput = isRepairableToolOutput(item.output);
+      // A known orphan call is still useful as a labeled user message even when its output is
+      // incomplete. With no call id and no output, preserve the invalid item so validation fails
+      // closed rather than pretending any tool result exists.
+      const knownNullOutput = callId.length > 0 && item.output == null;
+      if (!paired && (knownNullOutput || usableOutput)) {
         changed = true;
         repaired.push({
           type: "message",
           role: "user",
-          content: [{ type: "input_text", text: `[tool output for ${callId || "unknown call"}]\n${toolOutputText(item.output)}` }],
+          content: orphanedToolOutputContent(item.output, callId),
         });
         continue;
       }
@@ -2272,11 +2358,14 @@ export function createResponsesPassthroughAdapter(provider: OcxProviderConfig):
       // Same predicate as the routedCompaction gate in handleResponses(): an authMode check would
       // let a noncanonical custom forward provider skip this rewrite while the server still routes
       // it as a summarizer turn (#422). The compaction body build removes the tool surface and must
-      // therefore be the last routed transform: anything before it may depend on the declarations;
-      // anything after it cannot.
+      // therefore be the last routed transform that may depend on those declarations. Structural
+      // sanitizers below can still run after it.
       if (parsed._compactionRequest === true && !isCanonicalOpenAiForwardProvider(provider)) {
         outBody = buildRoutedCompactionBody(outBody);
       }
+      // Run after routed compaction so nested input_image parts are replaced before a malformed
+      // tool output is flattened to text and can no longer be inspected structurally.
+      outBody = repairUnidentifiedToolOutputItems(outBody);
       const threadServingIdentityChanged = parsed._stripReasoningEncryptedContent === true;
       const sanitizedBody = normalizeToolSchemas(
         stripSparkCompatibility(
diff --git a/tests/openai-responses-passthrough.test.ts b/tests/openai-responses-passthrough.test.ts
index 2048ed92e5..d411614626 100644
--- a/tests/openai-responses-passthrough.test.ts
+++ b/tests/openai-responses-passthrough.test.ts
@@ -2245,8 +2245,137 @@ describe("OpenAI Responses passthrough sanitization", () => {
     expect(rawDeltaBody.input).toHaveLength(1);
   });
 
+  test("api-key mode preserves delegated tool output text when call_id is missing", () => {
+    const adapter = createResponsesPassthroughAdapter({
+      adapter: "openai-responses",
+      baseUrl: "https://api.x.ai/v1",
+      authMode: "key" as const,
+      apiKey: "xai-test",
+    });
+
+    const body = JSON.parse(adapter.buildRequest({
+      ...parsedBase,
+      previousResponseId: undefined,
+      _rawBody: {
+        model: "grok-4.6",
+        input: [
+          {
+            type: "function_call_output",
+            id: "fco_delegation",
+            output: "<codex_delegation>Inspect the adapter.</codex_delegation>",
+          },
+        ],
+      },
+    }, meta).body) as { input: Record<string, unknown>[] };
+
+    expect(body.input).toEqual([{
+      type: "message",
+      role: "user",
+      content: [{
+        type: "input_text",
+        text: "[tool output for unknown call]\n<codex_delegation>Inspect the adapter.</codex_delegation>",
+      }],
+    }]);
+  });
+
+  test("api-key mode keeps stateful tool outputs with call_id intact", () => {
+    const adapter = createResponsesPassthroughAdapter({
+      adapter: "openai-responses",
+      baseUrl: "https://api.x.ai/v1",
+      authMode: "key" as const,
+      apiKey: "xai-test",
+    });
+    const statefulOutput = {
+      type: "function_call_output",
+      call_id: "call_from_previous_response",
+      output: "tool result",
+    };
+
+    const body = JSON.parse(adapter.buildRequest({
+      ...parsedBase,
+      _rawBody: {
+        model: "grok-4.6",
+        previous_response_id: "resp_stateful",
+        input: [statefulOutput],
+      },
+    }, meta).body) as { previous_response_id?: string; input: Record<string, unknown>[] };
+
+    expect(body.previous_response_id).toBe("resp_stateful");
+    expect(body.input).toEqual([statefulOutput]);
+  });
+
+  test("api-key mode preserves images when repairing output without call_id", () => {
+    const adapter = createResponsesPassthroughAdapter({
+      adapter: "openai-responses",
+      baseUrl: "https://api.x.ai/v1",
+      authMode: "key" as const,
+      apiKey: "xai-test",
+    });
+    const image = {
+      type: "input_image",
+      image_url: "data:image/png;base64,AAAA",
+      detail: "high",
+    };
+
+    const body = JSON.parse(adapter.buildRequest({
+      ...parsedBase,
+      _rawBody: {
+        model: "grok-4.6",
+        input: [{
+          type: "function_call_output",
+          output: [
+            { type: "input_text", text: "screenshot" },
+            image,
+            { type: "encrypted_content", encrypted_content: "opaque-tool-state" },
+          ],
+        }],
+      },
+    }, meta).body) as { input: Array<{ content: Record<string, unknown>[] }> };
+
+    expect(body.input[0]?.content).toEqual([
+      { type: "input_text", text: "[tool output for unknown call]" },
+      { type: "input_text", text: "screenshot" },
+      image,
+      { type: "input_text", text: "[encrypted content omitted]" },
+    ]);
+  });
+
+  test("api-key mode leaves invalid output without call_id fail-closed", () => {
+    const adapter = createResponsesPassthroughAdapter({
+      adapter: "openai-responses",
+      baseUrl: "https://api.x.ai/v1",
+      authMode: "key" as const,
+      apiKey: "xai-test",
+    });
+    const invalidOutputs = [
+      { type: "custom_tool_call_output" },
+      { type: "function_call_output", output: [{ type: "bogus", value: "not a tool-output part" }] },
+      {
+        type: "function_call_output",
+        output: [{ type: "input_image", image_url: "data:image/png;base64,AAAA", detail: ["high"] }],
+      },
+      {
+        type: "function_call_output",
+        output: [{ type: "input_image", image_url: 42, file_id: "file_1" }],
+      },
+      { type: "function_call_output", output: [{ type: "input_image", image_url: "" }] },
+      {
+        type: "function_call_output",
+        output: [{ type: { toString: null, valueOf: null }, text: "must not coerce" }],
+      },
+    ];
+
+    const body = JSON.parse(adapter.buildRequest({
+      ...parsedBase,
+      _rawBody: { model: "grok-4.6", input: invalidOutputs },
+    }, meta).body) as { input: Record<string, unknown>[] };
+
+    expect(body.input).toEqual(invalidOutputs);
+  });
+
   test("forward unexpanded miss converts orphan tool outputs and drops reasoning", () => {
     const adapter = createResponsesPassthroughAdapter(provider);
+    const image = { type: "input_image", image_url: "data:image/png;base64,AAAA" };
     const body = JSON.parse(adapter.buildRequest({
       ...parsedBase,
       _rawBody: {
@@ -2255,7 +2384,11 @@ describe("OpenAI Responses passthrough sanitization", () => {
         input: [
           { type: "reasoning", id: "rs_1", summary: [] },
           { type: "function_call_output", call_id: "call_orphan", output: "tool said hi" },
-          { type: "custom_tool_call_output", call_id: "call_custom", output: [{ type: "output_text", text: "custom out" }] },
+          {
+            type: "custom_tool_call_output",
+            call_id: "call_custom",
+            output: [{ type: "output_text", text: "custom out" }, image],
+          },
           { role: "user", content: "next question" },
         ],
       },
@@ -2272,7 +2405,11 @@ describe("OpenAI Responses passthrough sanitization", () => {
     expect(body.input[1]).toMatchObject({
       type: "message",
       role: "user",
-      content: [{ type: "input_text", text: "[tool output for call_custom]\ncustom out" }],
+      content: [
+        { type: "input_text", text: "[tool output for call_custom]" },
+        { type: "input_text", text: "custom out" },
+        image,
+      ],
     });
     expect(body.input[2]).toMatchObject({ role: "user", content: "next question" });
   });