fix: increase image tool maxTokens from 512 to 4096 (#11770)

* increase image tool maxTokens from 512 to 4096

* fix: cap image tool tokens by model capability (#11770) (thanks @detecti1)

* docs: fix changelog attribution for #11770

---------

Co-authored-by: Peter Steinberger <steipete@gmail.com>

CHANGELOG.md

@@ -10,6 +10,7 @@ Docs: https://docs.openclaw.ai
 
 ### Fixes
 
+- Agents/Image tool: cap image-analysis completion `maxTokens` by model capability (`min(4096, model.maxTokens)`) to avoid over-limit provider failures while still preventing truncation. (#11770) Thanks @detecti1.
 - Security/Canvas: serve A2UI assets via the shared safe-open path (`openFileWithinRoot`) to close traversal/TOCTOU gaps, with traversal and symlink regression coverage. (#10525) Thanks @abdelsfane.
 - Security/Gateway: breaking default-behavior change - canvas IP-based auth fallback now only accepts machine-scoped addresses (RFC1918, link-local, ULA IPv6, CGNAT); public-source IP matches now require bearer token auth. (#14661) Thanks @sumleo.
 - Security/WhatsApp: enforce `0o600` on `creds.json` and `creds.json.bak` on save/backup/restore paths to reduce credential file exposure. (#10529) Thanks @abdelsfane.

src/agents/tools/image-tool.e2e.test.ts

@@ -346,6 +346,18 @@ describe("image tool MiniMax VLM routing", () => {
 });
 
 describe("image tool response validation", () => {
+  it("caps image-tool max tokens by model capability", () => {
+    expect(__testing.resolveImageToolMaxTokens(4000)).toBe(4000);
+  });
+
+  it("keeps requested image-tool max tokens when model capability is higher", () => {
+    expect(__testing.resolveImageToolMaxTokens(8192)).toBe(4096);
+  });
+
+  it("falls back to requested image-tool max tokens when model capability is missing", () => {
+    expect(__testing.resolveImageToolMaxTokens(undefined)).toBe(4096);
+  });
+
   it("rejects image-model responses with no final text", () => {
     expect(() =>
       __testing.coerceImageAssistantText({

src/agents/tools/image-tool.ts

@@ -29,8 +29,20 @@ const ANTHROPIC_IMAGE_FALLBACK = "anthropic/claude-opus-4-5";
 export const __testing = {
   decodeDataUrl,
   coerceImageAssistantText,
+  resolveImageToolMaxTokens,
 } as const;
 
+function resolveImageToolMaxTokens(modelMaxTokens: number | undefined, requestedMaxTokens = 4096) {
+  if (
+    typeof modelMaxTokens !== "number" ||
+    !Number.isFinite(modelMaxTokens) ||
+    modelMaxTokens <= 0
+  ) {
+    return requestedMaxTokens;
+  }
+  return Math.min(requestedMaxTokens, modelMaxTokens);
+}
+
 function resolveDefaultModelRef(cfg?: OpenClawConfig): {
   provider: string;
   model: string;
@@ -287,7 +299,7 @@ async function runImagePrompt(params: {
       const context = buildImageContext(params.prompt, params.base64, params.mimeType);
       const message = await complete(model, context, {
         apiKey,
-        maxTokens: 512,
+        maxTokens: resolveImageToolMaxTokens(model.maxTokens),
       });
       const text = coerceImageAssistantText({
         message,

src/hooks/config.ts

@@ -70,12 +70,7 @@ export function hasBinary(bin: string): boolean {
   const parts = pathEnv.split(path.delimiter).filter(Boolean);
   const extensions =
     process.platform === "win32"
-      ? [
+      ? ["", ...(process.env.PATHEXT ?? ".EXE;.CMD;.BAT;.COM").split(";").filter(Boolean)]
-          "",
-          ...(process.env.PATHEXT ?? ".EXE;.CMD;.BAT;.COM")
-            .split(";")
-            .filter(Boolean),
-        ]
       : [""];
   for (const part of parts) {
     for (const ext of extensions) {