fix(token-limit): normalize detection across providers

Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)

Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>

src/hooks/todo-continuation-enforcer/token-limit-detection.ts

@@ -1,8 +1,6 @@
-const TOKEN_LIMIT_ERROR_NAMES = new Set([
-  "contextlengtherror",
-])
+import { isRetryableModelError } from "../../shared/model-error-classifier"
 
-const TOKEN_LIMIT_KEYWORDS = [
+const TOKEN_LIMIT_FALLBACK_PATTERNS = [
   "prompt is too long",
   "is too long",
   "context_length_exceeded",
@@ -11,16 +9,29 @@ const TOKEN_LIMIT_KEYWORDS = [
   "too many tokens",
 ]
 
+const TOKEN_LIMIT_ERROR_NAMES = new Set([
+  "contextlengtherror",
+  "context_length_exceeded",
+])
+
 export function isTokenLimitError(error: { name?: string; message?: string } | undefined): boolean {
   if (!error) return false
 
-  if (error.name && TOKEN_LIMIT_ERROR_NAMES.has(error.name.toLowerCase())) {
-    return true
+  const isRetryable = isRetryableModelError({
+    name: error.name,
+    message: error.message,
+  })
+
+  if (!isRetryable && error.name) {
+    const errorNameLower = error.name.toLowerCase()
+    if (TOKEN_LIMIT_ERROR_NAMES.has(errorNameLower)) {
+      return true
+    }
   }
 
   if (error.message) {
     const lower = error.message.toLowerCase()
-    return TOKEN_LIMIT_KEYWORDS.some((keyword) => lower.includes(keyword))
+    return TOKEN_LIMIT_FALLBACK_PATTERNS.some((pattern) => lower.includes(pattern))
   }
 
   return false