fix(google): preserve matched and orphan Vertex Argon tool images

docs-site/src/content/docs/guides/providers.md

@@ -190,6 +190,13 @@ eligible Enterprise accounts. Cloud Code Assist serves Argon only to the Antigra
 OpenCodex sends Argon requests and model discovery with the CLI User-Agent
 (`antigravity/cli/1.3.4`) and keeps the IDE User-Agent for every other model.
 
+On Vertex AI, `gemini-4-argon` receives images returned by tools, such as screenshots, inside
+the matching function response. This avoids the upstream 400 error "Requests ending with a
+model turn are not supported" after an image-producing tool call while preserving the images.
+Images from duplicate results or results without a matching call are preserved in a separate
+message after the tool responses, with a marker explaining that they could not be paired.
+No configuration change is needed; this handling is specific to Vertex Argon.
+
 For Gemini on Antigravity, including Argon, provider-marked thought summaries stay separate
 from the answer: Responses uses reasoning summary items, Chat Completions uses
 `reasoning_content`, and Messages uses `thinking` blocks. Clients decide how to display those

docs-site/src/content/docs/ko/guides/providers.md

@@ -121,6 +121,13 @@ ocx logout <provider>
 
 Google Antigravity 계정·제공자 할당량 확인은 모델 목록 폴백을 포함해 고정된 Google 회계 엔드포인트를 사용합니다. 해당 목적지의 투명 Fake-IP DNS를 지원하며 TLS 검증, 리다이렉트 거부, 사설 주소 검사는 유지합니다. 사용자 지정 base URL은 모델 요청에만 적용되며 할당량 목적지는 바꾸지 않습니다. `NO_PROXY`는 기존 직접 연결 정책을 유지합니다.
 
+Vertex AI의 `gemini-4-argon`에는 스크린샷 같은 도구 결과 이미지를 해당 함수 응답 안에 담아
+전송합니다. 이미지가 있는 도구 호출 뒤에 발생하는 400 오류
+"Requests ending with a model turn are not supported"를 방지하며 이미지를 보존합니다.
+중복된 결과나 일치하는 호출이 없는 결과의 이미지는 도구 응답 뒤의 별도 메시지로 보존하고,
+호출과 연결할 수 없었다는 표시를 남깁니다.
+별도의 설정 변경은 필요하지 않으며, 이 처리는 Vertex Argon에만 적용됩니다.
+
 ### Google 도구 스키마 손실 진단
 
 Google 도구 선언은 선택된 엔드포인트 클래스에 맞춰 컴파일됩니다. `ocx debug provider on`,

src/adapters/google.ts

@@ -333,13 +333,20 @@ function geminiToolResultParts(
   msg: OcxToolResultMessage,
   wireName: string,
   wireCallId: string,
+  nestToolResultImages: boolean,
 ): unknown[] {
   const functionResponse: Record<string, unknown> = {
     name: wireName,
     response: { result: geminiToolResultText(msg.content) },
     id: wireCallId,
   };
-  return [{ functionResponse }, ...toolResultImageParts(msg.content)];
+  const images = toolResultImageParts(msg.content);
+  // Vertex Argon rejects image siblings beside a functionResponse as an unsupported model tail.
+  if (nestToolResultImages && images.length > 0) {
+    functionResponse.parts = images;
+    return [{ functionResponse }];
+  }
+  return [{ functionResponse }, ...images];
 }
 
 function geminiMissingToolResultPart(wireName: string, wireCallId: string): unknown {
@@ -370,6 +377,7 @@ function messagesToGeminiFormat(
   identityModelId: string,
   stripRejectedClaudeSdkParagraph = false,
   isCloudCodeAssist = false,
+  nestToolResultImages = false,
 ): { systemInstruction?: unknown; contents: unknown[]; replayedCallIds: string[] } {
   // Neutralize Codex's GPT-5 identity line (Gemini/Antigravity share this path) so a routed model
   // never misreports as GPT-5/OpenAI, and never leaks the proxy identity upstream.
@@ -551,13 +559,17 @@ function messagesToGeminiFormat(
           const responseParts: unknown[] = [];
           for (const call of toolCalls) {
             const result = resultsById.get(call.wireCallId);
-            if (result) responseParts.push(...geminiToolResultParts(result, call.wireName, call.wireCallId));
+            if (result) responseParts.push(...geminiToolResultParts(result, call.wireName, call.wireCallId, nestToolResultImages));
             else responseParts.push(geminiMissingToolResultPart(call.wireName, call.wireCallId));
           }
-          for (const orphan of orphanResults) {
-            responseParts.push(...geminiOrphanToolResultParts(orphan));
-          }
+          const orphanParts = orphanResults.flatMap(geminiOrphanToolResultParts);
           contents.push({ role: "user", parts: responseParts });
+          // Orphan images cannot share Argon's response batch or claim another call's ownership.
+          if (nestToolResultImages && orphanParts.some(part => typeof part === "object" && part !== null && "inline_data" in part)) {
+            contents.push({ role: "user", parts: orphanParts });
+          } else {
+            responseParts.push(...orphanParts);
+          }
           i = j - 1;
         }
         break;
@@ -1026,6 +1038,7 @@ export function createGoogleAdapter(provider: OcxProviderConfig): ProviderAdapte
         identityModelId,
         stripRejectedClaudeSdkParagraph,
         isCloudCodeAssist,
+        provider.googleMode === "vertex" && parsed.modelId === "gemini-4-argon",
       );
       lastInjectedCallIds = [...replayedCallIds];
       lastReasoningReplayScope = parsed._reasoningReplayScope;

structure/decisions/ADR-7279-vertex-argon-tool-images.md

@@ -0,0 +1,16 @@
+# ADR-7279: decision recorded under "Google tool-result adjacency repair"
+
+- Contract owner: [providers/google.md](../providers/google.md#google-tool-result-adjacency-repair)
+- Purpose: allow Vertex `gemini-4-argon` to continue after a tool returns an image (#7279).
+- Prior constraint: tool images were sibling `inline_data` parts beside `functionResponse`.
+  A synthetic live request ending in `user` still received HTTP 400, "Requests ending with
+  a model turn are not supported"; adding final user text did not repair it.
+- Alternatives: discard images, move them to a separate user turn, or nest them inside the
+  response. Discarding loses information; separate turns obscure ownership in parallel results.
+- Choice: nest matched result images in `functionResponse.parts` only for Vertex
+  `gemini-4-argon`, preserving the existing text, IDs, names, order and image bytes. The nested
+  synthetic request succeeded and reported IMAGE input tokens. Other routes retain their existing
+  representation because this incompatibility was reproduced only on Vertex Argon.
+- Consequences: no new configuration, retry or shared-history mutation. Orphan results still
+  degrade to marked text and sibling images; existing adjacency and thought-signature repairs
+  remain intact. Revisit the model scope when another route has equivalent upstream evidence.

structure/decisions/ADR-7280-vertex-argon-orphan-images.md

@@ -0,0 +1,15 @@
+# ADR-7280: decision recorded under "Google tool-result adjacency repair"
+
+- Contract owner: [providers/google.md](../providers/google.md#google-tool-result-adjacency-repair)
+- Purpose: preserve adjacent orphan images without reopening the Vertex Argon 400 from #7279.
+- Prior constraint: ADR-7279 nested matched images, but duplicate or mismatched result images
+  still shared their user turn with the function responses. Both cases reproduced the same
+  upstream 400; moving the orphan content to a following user turn succeeded.
+- Alternatives: discard orphan images, attach them to a matched response, or preserve them in
+  a separate turn. The first loses content; the second invents ownership.
+- Choice: for Vertex `gemini-4-argon` only, move the entire adjacent orphan batch to a following
+  user turn when it contains a representable image. Keep the complete response batch adjacent
+  to its calls and preserve the orphan markers, image bytes, and orphan result order.
+- Consequences: no history mutation, invented success, configuration or retry. Text-only and
+  standalone orphans keep their prior shape; other Google routes are unchanged. The existing
+  Vertex regression file covers duplicates, mismatches, missing results and route isolation.

structure/providers/google.md

@@ -132,9 +132,18 @@ Google-family requests serialize a model tool-call turn and its results as one a
 `model -> user` pair. The user turn contains exactly one `functionResponse` for every representable
 call in original call order. Missing results use an explicit unknown-history marker; duplicate,
 mismatched, and standalone results become marked text instead of unpaired function responses.
-Representable data-URL images remain sibling `inline_data` parts in either case.
+For Vertex `gemini-4-argon`, matched results carry representable data-URL images inside
+`functionResponse.parts`, preserving each image's call ownership. Sibling images on that route can
+produce a misleading "Requests ending with a model turn are not supported" error even when the
+last turn is `user`. When an adjacent orphan batch contains representable images on that route,
+the entire marked orphan batch follows the complete function-response batch in a separate user
+turn, preserving orphan result order and image bytes without assigning an orphan image to another call. Text-only
+orphans, standalone orphans and other routes retain their existing representation.
+`tests/adapters/google/google-vertex-thought-signature.test.ts` covers the scoped wire format.
 
 > Decision record: [ADR-0058](../decisions/ADR-0058-google-tool-result-adjacency-repair.md)
+> Decision record: [ADR-7279](../decisions/ADR-7279-vertex-argon-tool-images.md)
+> Decision record: [ADR-7280](../decisions/ADR-7280-vertex-argon-orphan-images.md)
 
 ## Google opening functionCall repair