fix(translator): don't drop image-only user messages in prepareClaudeRequest

hasValidContent() only treated text/tool_use/tool_result blocks as valid
content, so a user message containing only an image block was filtered
out as empty. When it was the only non-system message, this left an
empty messages array and Anthropic rejected the request.

open-sse/translator/formats/claude.js

@@ -16,7 +16,9 @@ export function hasValidContent(msg) {
     return msg.content.some(block =>
       (block.type === CLAUDE_BLOCK.TEXT && block.text?.trim()) ||
       block.type === CLAUDE_BLOCK.TOOL_USE ||
-      block.type === CLAUDE_BLOCK.TOOL_RESULT
+      block.type === CLAUDE_BLOCK.TOOL_RESULT ||
+      block.type === CLAUDE_BLOCK.IMAGE ||
+      block.type === CLAUDE_BLOCK.DOCUMENT
     );
   }
   return false;

tests/translator/bugs-toClaude-context.test.js

@@ -67,6 +67,27 @@ describe("OpenAI → Claude context mapping", () => {
     expect(JSON.stringify(out), "remote image dropped").toContain("pic.png");
   });
 
+  // claude.js hasValidContent() — a user message whose content is only an
+  // image (no text block) was filtered out by prepareClaudeRequest's
+  // "drop empty messages" pass, so an image-only turn (e.g. a vision
+  // describe request with no accompanying prompt text in the user message)
+  // produced an empty `messages` array and Anthropic rejected the request
+  // with "messages: at least one message is required".
+  it("user message with only an image is not dropped as empty", () => {
+    const out = T({
+      messages: [
+        { role: "system", content: "Describe the image." },
+        { role: "user", content: [
+          { type: "image_url", image_url: { url: "data:image/png;base64,AAAA" } },
+        ] },
+      ],
+    });
+    expect(out.messages.length, "image-only user message was dropped").toBeGreaterThan(0);
+    expect(out.messages[0].content).toEqual(
+      expect.arrayContaining([expect.objectContaining({ type: "image" })])
+    );
+  });
+
   // prepareClaudeRequest reconciles max_tokens vs thinking.budget_tokens.
   // applyThinking runs after adjustMaxTokens caps max_tokens, so a claude-budget
   // model at "max" effort (budget 128000) can exceed the clamped max_tokens and