fix(embedded): classify model_context_window_exceeded as context overflow, trigger compaction (#35934)

Merged via squash.

Prepared head SHA: 20fa77289c
Co-authored-by: RealKai42 <44634134+RealKai42@users.noreply.github.com>
Co-authored-by: jalehman <550978+jalehman@users.noreply.github.com>
Reviewed-by: @jalehman

src/agents/pi-embedded-helpers.isbillingerrormessage.test.ts

@@ -269,6 +269,21 @@ describe("isContextOverflowError", () => {
     }
   });
 
+  it("matches model_context_window_exceeded stop reason surfaced by pi-ai", () => {
+    // Anthropic API (and some OpenAI-compatible providers like ZhipuAI/GLM) return
+    // stop_reason: "model_context_window_exceeded" when the context window is hit.
+    // The pi-ai library surfaces this as "Unhandled stop reason: model_context_window_exceeded".
+    const samples = [
+      "Unhandled stop reason: model_context_window_exceeded",
+      "model_context_window_exceeded",
+      "context_window_exceeded",
+      "Unhandled stop reason: context_window_exceeded",
+    ];
+    for (const sample of samples) {
+      expect(isContextOverflowError(sample)).toBe(true);
+    }
+  });
+
   it("matches Chinese context overflow error messages from proxy providers", () => {
     const samples = [
       "上下文过长",

src/agents/pi-embedded-helpers/errors.ts

@@ -105,6 +105,9 @@ export function isContextOverflowError(errorMessage?: string): boolean {
     (lower.includes("max_tokens") && lower.includes("exceed") && lower.includes("context")) ||
     (lower.includes("input length") && lower.includes("exceed") && lower.includes("context")) ||
     (lower.includes("413") && lower.includes("too large")) ||
+    // Anthropic API and OpenAI-compatible providers (e.g. ZhipuAI/GLM) return this stop reason
+    // when the context window is exceeded. pi-ai surfaces it as "Unhandled stop reason: model_context_window_exceeded".
+    lower.includes("context_window_exceeded") ||
     // Chinese proxy error messages for context overflow
     errorMessage.includes("上下文过长") ||
     errorMessage.includes("上下文超出") ||

src/agents/pi-embedded-runner/model.test.ts

@@ -278,6 +278,118 @@ describe("resolveModel", () => {
     expect(result.model?.reasoning).toBe(true);
   });
 
+  it("prefers configured provider api metadata over discovered registry model", () => {
+    mockDiscoveredModel({
+      provider: "onehub",
+      modelId: "glm-5",
+      templateModel: {
+        id: "glm-5",
+        name: "GLM-5 (cached)",
+        provider: "onehub",
+        api: "anthropic-messages",
+        baseUrl: "https://old-provider.example.com/v1",
+        reasoning: false,
+        input: ["text"],
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+        contextWindow: 8192,
+        maxTokens: 2048,
+      },
+    });
+
+    const cfg = {
+      models: {
+        providers: {
+          onehub: {
+            baseUrl: "http://new-provider.example.com/v1",
+            api: "openai-completions",
+            models: [
+              {
+                ...makeModel("glm-5"),
+                api: "openai-completions",
+                reasoning: true,
+                contextWindow: 198000,
+                maxTokens: 16000,
+              },
+            ],
+          },
+        },
+      },
+    } as OpenClawConfig;
+
+    const result = resolveModel("onehub", "glm-5", "/tmp/agent", cfg);
+
+    expect(result.error).toBeUndefined();
+    expect(result.model).toMatchObject({
+      provider: "onehub",
+      id: "glm-5",
+      api: "openai-completions",
+      baseUrl: "http://new-provider.example.com/v1",
+      reasoning: true,
+      contextWindow: 198000,
+      maxTokens: 16000,
+    });
+  });
+
+  it("prefers exact provider config over normalized alias match when both keys exist", () => {
+    mockDiscoveredModel({
+      provider: "qwen",
+      modelId: "qwen3-coder-plus",
+      templateModel: {
+        id: "qwen3-coder-plus",
+        name: "Qwen3 Coder Plus",
+        provider: "qwen",
+        api: "openai-completions",
+        baseUrl: "https://default-provider.example.com/v1",
+        reasoning: false,
+        input: ["text"],
+        cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+        contextWindow: 8192,
+        maxTokens: 2048,
+      },
+    });
+
+    const cfg = {
+      models: {
+        providers: {
+          "qwen-portal": {
+            baseUrl: "https://canonical-provider.example.com/v1",
+            api: "openai-completions",
+            headers: { "X-Provider": "canonical" },
+            models: [{ ...makeModel("qwen3-coder-plus"), reasoning: false }],
+          },
+          qwen: {
+            baseUrl: "https://alias-provider.example.com/v1",
+            api: "anthropic-messages",
+            headers: { "X-Provider": "alias" },
+            models: [
+              {
+                ...makeModel("qwen3-coder-plus"),
+                api: "anthropic-messages",
+                reasoning: true,
+                contextWindow: 262144,
+                maxTokens: 32768,
+              },
+            ],
+          },
+        },
+      },
+    } as OpenClawConfig;
+
+    const result = resolveModel("qwen", "qwen3-coder-plus", "/tmp/agent", cfg);
+
+    expect(result.error).toBeUndefined();
+    expect(result.model).toMatchObject({
+      provider: "qwen",
+      id: "qwen3-coder-plus",
+      api: "anthropic-messages",
+      baseUrl: "https://alias-provider.example.com",
+      reasoning: true,
+      contextWindow: 262144,
+      maxTokens: 32768,
+      headers: { "X-Provider": "alias" },
+    });
+  });
+
   it("builds an openai-codex fallback for gpt-5.3-codex", () => {
     mockOpenAICodexTemplateModel();
 

src/agents/pi-embedded-runner/model.ts

@@ -7,7 +7,7 @@ import { DEFAULT_CONTEXT_TOKENS } from "../defaults.js";
 import { buildModelAliasLines } from "../model-alias-lines.js";
 import { normalizeModelCompat } from "../model-compat.js";
 import { resolveForwardCompatModel } from "../model-forward-compat.js";
-import { normalizeProviderId } from "../model-selection.js";
+import { findNormalizedProviderValue, normalizeProviderId } from "../model-selection.js";
 import { discoverAuthStorage, discoverModels } from "../pi-model-discovery.js";
 
 type InlineModelEntry = ModelDefinitionConfig & {
@@ -24,6 +24,60 @@ type InlineProviderConfig = {
 
 export { buildModelAliasLines };
 
+function resolveConfiguredProviderConfig(
+  cfg: OpenClawConfig | undefined,
+  provider: string,
+): InlineProviderConfig | undefined {
+  const configuredProviders = cfg?.models?.providers;
+  if (!configuredProviders) {
+    return undefined;
+  }
+  const exactProviderConfig = configuredProviders[provider];
+  if (exactProviderConfig) {
+    return exactProviderConfig;
+  }
+  return findNormalizedProviderValue(configuredProviders, provider);
+}
+
+function applyConfiguredProviderOverrides(params: {
+  discoveredModel: Model<Api>;
+  providerConfig?: InlineProviderConfig;
+  modelId: string;
+}): Model<Api> {
+  const { discoveredModel, providerConfig, modelId } = params;
+  if (!providerConfig) {
+    return discoveredModel;
+  }
+  const configuredModel = providerConfig.models?.find((candidate) => candidate.id === modelId);
+  if (
+    !configuredModel &&
+    !providerConfig.baseUrl &&
+    !providerConfig.api &&
+    !providerConfig.headers
+  ) {
+    return discoveredModel;
+  }
+  return {
+    ...discoveredModel,
+    api: configuredModel?.api ?? providerConfig.api ?? discoveredModel.api,
+    baseUrl: providerConfig.baseUrl ?? discoveredModel.baseUrl,
+    reasoning: configuredModel?.reasoning ?? discoveredModel.reasoning,
+    input: configuredModel?.input ?? discoveredModel.input,
+    cost: configuredModel?.cost ?? discoveredModel.cost,
+    contextWindow: configuredModel?.contextWindow ?? discoveredModel.contextWindow,
+    maxTokens: configuredModel?.maxTokens ?? discoveredModel.maxTokens,
+    headers:
+      providerConfig.headers || configuredModel?.headers
+        ? {
+            ...discoveredModel.headers,
+            ...providerConfig.headers,
+            ...configuredModel?.headers,
+          }
+        : discoveredModel.headers,
+    compat: configuredModel?.compat ?? discoveredModel.compat,
+  };
+}
+
 export function buildInlineProviderModels(
   providers: Record<string, InlineProviderConfig>,
 ): InlineModelEntry[] {
@@ -59,6 +113,7 @@ export function resolveModel(
   const resolvedAgentDir = agentDir ?? resolveOpenClawAgentDir();
   const authStorage = discoverAuthStorage(resolvedAgentDir);
   const modelRegistry = discoverModels(authStorage, resolvedAgentDir);
+  const providerConfig = resolveConfiguredProviderConfig(cfg, provider);
   const model = modelRegistry.find(provider, modelId) as Model<Api> | null;
 
   if (!model) {
@@ -100,7 +155,7 @@ export function resolveModel(
       } as Model<Api>);
       return { model: fallbackModel, authStorage, modelRegistry };
     }
-    const providerCfg = providers[provider];
+    const providerCfg = providerConfig;
     if (providerCfg || modelId.startsWith("mock-")) {
       const configuredModel = providerCfg?.models?.find((candidate) => candidate.id === modelId);
       const fallbackModel: Model<Api> = normalizeModelCompat({
@@ -133,21 +188,17 @@ export function resolveModel(
       modelRegistry,
     };
   }
-  const providerOverride = cfg?.models?.providers?.[provider] as InlineProviderConfig | undefined;
-  if (providerOverride?.baseUrl || providerOverride?.headers) {
-    const overridden: Model<Api> & { headers?: Record<string, string> } = { ...model };
-    if (providerOverride.baseUrl) {
-      overridden.baseUrl = providerOverride.baseUrl;
-    }
-    if (providerOverride.headers) {
-      overridden.headers = {
-        ...(model as Model<Api> & { headers?: Record<string, string> }).headers,
-        ...providerOverride.headers,
-      };
-    }
-    return { model: normalizeModelCompat(overridden), authStorage, modelRegistry };
-  }
-  return { model: normalizeModelCompat(model), authStorage, modelRegistry };
+  return {
+    model: normalizeModelCompat(
+      applyConfiguredProviderOverrides({
+        discoveredModel: model,
+        providerConfig,
+        modelId,
+      }),
+    ),
+    authStorage,
+    modelRegistry,
+  };
 }
 
 /**