default model glm-5.2; Claude Code sets top-level model to the Hanzo default

src/lib/models.ts

@@ -169,9 +169,8 @@ export function resolveModel(input: string): string {
 }
 
 /**
- * Recommended default: `zen-auto`. The cloud routes it to the best available
- * model, so it keeps working no matter how the catalog changes — the most
- * "dynamic, works no matter what" choice for a fresh setup. Users can pick any
- * concrete id in the wizard instead.
+ * Recommended default: `glm-5.2`. It is the stable concrete default the helper
+ * writes for fresh setups; users can still override it with any effort word or
+ * explicit model id in the wizard.
  */
-export const DEFAULT_MODEL = 'zen-auto';
+export const DEFAULT_MODEL = 'glm-5.2';

src/targets/claude-code.ts

@@ -3,8 +3,9 @@
  *
  * Uses the `providers.hanzo` block in ~/.claude/settings.json — an additive,
  * non-destructive entry that lets users switch to Hanzo models per-session
- * (`/model hanzo/deepseek-v4-pro`) without replacing the default Anthropic
- * connection. The env block is NOT touched so the current session stays intact.
+ * without replacing the default connection. The top-level `model` field is
+ * updated to the selected Hanzo default, while the rest of the settings stay
+ * intact.
  *
  * Config paths are identical on all platforms (home-relative). Claude Code
  * also looks at CLAUDE_SETTINGS_PATH env override.
@@ -43,10 +44,10 @@ export const claudeCode: CodingTarget = {
         name: 'Hanzo AI',
         apiKey: creds.apiKey,
         baseURL: creds.apiBase + '/v1',
-        // Live catalog from the cloud; fall back to just the default if offline.
         models: creds.models && creds.models.length > 0 ? creds.models : [creds.model],
       },
     };
+    settings.model = creds.model;
     writeJson(SETTINGS, settings);
 
     const claudeJson = readJson<{ hasCompletedOnboarding?: boolean }>(CLAUDE_JSON);