Enable Claude model discovery by default and remove the opt-in flag

README.md

@@ -51,6 +51,11 @@ On first launch of a model-backed agent, `ug` prompts for a Databricks
 workspace, authenticates, and writes local agent config. Later launches reuse
 the saved workspace and credentials.
 
+Without a managed workspace config, `ug claude` automatically discovers gateway
+models for Claude Code's `/model` picker. Discovery defaults to `system.ai` when
+no provider or model location is selected. Use `--provider` or `--model-location`
+to select another model source; managed workspace configs control their own sources.
+
 ## Configure
 
 ```bash

src/ucode/cli.py

@@ -2600,7 +2600,10 @@ def _launch_tool(
             # rewriting it; the admin's location exists only for this launch.
             provider = None
             parent_schema = managed_parent_schema
-        if tool == "claude" and (managed_provider or managed_parent_schema):
+        # Unmanaged Claude launches discover gateway models automatically; with no
+        # provider or parent header the gateway defaults to system.ai. Managed
+        # configs opt into discovery by selecting an MPS or Unity Catalog location.
+        if tool == "claude" and (managed is None or managed_provider or managed_parent_schema):
             os.environ[claude_agent.GATEWAY_MODEL_DISCOVERY_ENV_VAR] = "1"
         # The environment switch remains a developer override; managed config is the workspace
         # policy equivalent and must take effect before launch options are computed.
@@ -3151,14 +3154,6 @@ def claude_cmd(
         str | None,
         typer.Option("--scopes", hidden=True, help="Comma-separated custom OAuth scopes."),
     ] = None,
-    enable_model_discovery: Annotated[
-        bool,
-        typer.Option(
-            "--enable-model-discovery",
-            hidden=True,
-            help="Enable AI Gateway models in Claude Code's model picker.",
-        ),
-    ] = False,
     enable_smart_routing_flag: Annotated[
         bool,
         typer.Option(
@@ -3188,8 +3183,6 @@ def claude_cmd(
         claude_agent.disable_smart_routing(load_state())
         print_success("Claude Code smart routing disabled; ug routing hooks removed")
         return
-    if enable_model_discovery or model_location is not None or provider is not None:
-        os.environ[claude_agent.GATEWAY_MODEL_DISCOVERY_ENV_VAR] = "1"
     with _smart_routing_v2_flag(enable_smart_routing_flag):
         with _disable_smart_routing_for_subcommand("claude", ctx):
             _launch_tool(

tests/README.md

@@ -33,7 +33,7 @@ All tests live directly in `integration/`; shared mechanics live in `utils/`.
 
 | Test | User action | Expected evidence |
 | --- | --- | --- |
-| `test_ug_configure_claude_databricks` | Configure Databricks Hosted; execute the generated auth helper; open Claude TUI and read a file | Generated helper invokes `ug` with clean token stdout; assistant returns an unpredictable file value; normal exit; reopen with working keyboard input |
+| `test_ug_configure_claude_databricks` | Configure Databricks Hosted; execute the generated auth helper; launch plain `ug claude`, read a file, and open `/model` | Generated helper invokes `ug` with clean token stdout; assistant returns an unpredictable file value; native discovery caches `system.ai` models and the picker shows a discovered model without an opt-in flag; normal exit; reopen with working keyboard input |
 | `test_ug_configure_claude_anthropic_mps` | Select Anthropic MPS in the real configure picker; launch Claude | Saved provider in status; completed TUI file task; normal exit |
 | `test_ug_configure_codex_databricks` | Configure Databricks Hosted; execute the generated auth helper; open Codex TUI and read a file | Generated helper invokes `ug` with clean token stdout; completed assistant answer contains the file value; normal exit and reopen |
 | `test_ug_configure_codex_openai_mps` | Select OpenAI MPS in the real configure picker; launch Codex | Saved provider in status; completed TUI file task; normal exit |

tests/integration/README.md

@@ -80,6 +80,11 @@ bearer override and drive their real local web-search MCP handshake/tool listing
 These assert protocol stdout without stripping ANSI escapes and make no workspace
 requests. The live Hosted configure journeys additionally execute the actual `ug`
 auth helper written into each agent's configuration before completing a real TUI task.
+The Claude journey also opens `/model` after a plain `ug claude` launch, requires
+its native gateway cache to contain `system.ai` models, and checks that a discovered
+model appears in the picker. No managed config, provider, model location, discovery
+flag, or inherited discovery environment variable enables this path. This runs with
+the pinned Claude version (currently 2.1.268 in CI).
 
 ## Test layout and format
 

tests/integration/test_ug_configure_claude.py

@@ -13,10 +13,12 @@
 
 @pytest.mark.smoke
 def test_ug_configure_claude_databricks(live_session, workspace):
-    """Scenario: configure Claude with Databricks Hosted and use its TUI.
+    """Scenario: configure Databricks Hosted, launch plain ug claude, and open /model.
 
-    Expected: configure succeeds, Claude returns a file value through the real
-    gateway, exits normally, and can reopen the configuration ug created.
+    Expected: without managed config, a provider, a model location, or a discovery
+    flag, Claude caches system.ai models from the gateway and shows a discovered
+    model in its real picker. It returns a file value through the real gateway,
+    exits normally, and can reopen the configuration ug created.
     The generated auth helper uses ug and prints only the supplied bearer.
     Optional AI Tools are disabled; the selected agent version is kept pinned.
     """
@@ -49,6 +51,16 @@ def test_ug_configure_claude_databricks(live_session, workspace):
         tui.boot()
         tui.submit(task.prompt)
         tui.wait_for_task(task)
+        screen = tui.open_model_picker()
+        cache = json.loads((session.home / ".claude/cache/gateway-models.json").read_text())
+        assert cache["baseUrl"] == workspace.rstrip("/") + "/ai-gateway/anthropic", cache
+        models = cache["models"]
+        system_ai_models = [model for model in models if model["id"].startswith("system.ai.")]
+        assert system_ai_models, cache
+        assert any(
+            model["id"] in screen or (model.get("display_name") and model["display_name"] in screen)
+            for model in system_ai_models
+        ), screen
         tui.exit_normally()
     task.assert_completed(session, "claude")
     session.assert_not_routed()

tests/test_cli.py

@@ -729,30 +729,30 @@ def test_codex_forwarded_model_is_not_printed_in_launch_summary(
         assert "Model: system.ai.gpt-5-6-luna" not in output
         assert mock_launch.call_args.args[2] == forwarded_args
 
-    def test_claude_enable_model_discovery_sets_ucode_env(self):
-        with patch("ucode.cli._launch_tool") as mock_launch:
-            result = runner.invoke(app, ["claude", "--enable-model-discovery"])
+    def test_unmanaged_claude_launch_enables_model_discovery(self):
+        with _launch_policy_patches(None) as calls:
+            result = runner.invoke(app, ["claude"])
 
         assert result.exit_code == 0, result.output
         assert os.environ["ENABLE_CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY"] == "1"
-        assert mock_launch.call_args.args[1].args == []
+        calls["launch"].assert_called_once()
 
     def test_claude_model_location_is_forwarded(self):
-        with patch("ucode.cli._launch_tool") as mock_launch:
+        with _launch_policy_patches(None) as calls:
             result = runner.invoke(app, ["claude", "--model-location", "main.default"])
 
         assert result.exit_code == 0, result.output
-        assert mock_launch.call_args.kwargs["parent_schema"] == "main.default"
-        assert mock_launch.call_args.args[1].args == []
+        assert calls["configure"].call_args.kwargs["parent_schema"] == "main.default"
+        assert calls["launch"].call_args.args[2] == []
         assert os.environ["ENABLE_CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY"] == "1"
 
     def test_claude_provider_enables_model_discovery(self):
-        with patch("ucode.cli._launch_tool") as mock_launch:
+        with _launch_policy_patches(None) as calls:
             result = runner.invoke(app, ["claude", "--provider", "main.default.anthropic"])
 
         assert result.exit_code == 0, result.output
-        assert mock_launch.call_args.kwargs["provider"] == "main.default.anthropic"
-        assert mock_launch.call_args.args[1].args == []
+        assert calls["configure"].call_args.kwargs["provider"] == "main.default.anthropic"
+        assert calls["launch"].call_args.args[2] == []
         assert os.environ["ENABLE_CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY"] == "1"
 
     def test_codex_model_location_is_forwarded(self):
@@ -803,12 +803,6 @@ def test_invalid_model_location_is_rejected(self, tool):
         assert result.exit_code == 1
         assert "--model-location must be `<catalog>.<schema>`." in _strip_ansi(result.output)
 
-    def test_claude_enable_model_discovery_is_hidden_from_help(self):
-        result = runner.invoke(app, ["claude", "--help"])
-
-        assert result.exit_code == 0, result.output
-        assert "--enable-model-discovery" not in result.output
-
     def test_codex_disable_removes_hooks_without_launching(self):
         with (
             patch("ucode.cli.load_state", return_value=MINIMAL_STATE),
@@ -1027,6 +1021,19 @@ def test_persisted_provider_is_not_mistaken_for_an_explicit_option(self):
 
 
 class TestManagedClaudeModelDiscovery:
+    def test_managed_static_models_do_not_enable_discovery(self):
+        managed = {
+            "enabled_agents": {
+                "claude": {"model_config": {"model_services": ["system.ai.claude-sonnet-5"]}}
+            }
+        }
+        with _launch_policy_patches(managed) as calls:
+            result = runner.invoke(app, ["claude"])
+
+        assert result.exit_code == 0, result.output
+        calls["launch"].assert_called_once()
+        assert "ENABLE_CLAUDE_CODE_GATEWAY_MODEL_DISCOVERY" not in os.environ
+
     MPS_CONFIG = {
         "enabled_agents": {
             "claude": {