Add project tools and strict provider execution

tests/test_runner.py

@@ -229,6 +229,23 @@ class TestExecutePromptStep:
         assert success is False
         assert output == ""
 
+    @patch('cmdforge.runner.call_provider')
+    def test_step_can_deny_fallback_independently_of_caller(self, mock_call):
+        mock_call.return_value = ProviderResult(
+            text="private", success=True,
+        )
+        step = PromptStep(
+            prompt="Private", provider="local", output_var="out",
+            plain_text=True, fallback_policy="deny",
+        )
+
+        output, success = execute_prompt_step(step, {"input": ""})
+
+        assert success is True
+        assert output == "private"
+        policy = mock_call.call_args.kwargs["execution_policy"]
+        assert policy.allow_fallback is False
+
     @patch('cmdforge.runner.mock_provider')
     def test_mock_provider_used(self, mock_mock):
         mock_mock.return_value = ProviderResult(text="mock response", success=True)
@@ -988,6 +1005,58 @@ class TestCreateArgumentParser:
 
         assert args.count == "10"
 
+    def test_parser_enforces_typed_boolean_integer_and_enum_arguments(self):
+        tool = Tool(
+            name="typed",
+            arguments=[
+                ToolArgument(
+                    flag="--project", variable="project",
+                    type="boolean", default=False,
+                ),
+                ToolArgument(
+                    flag="--count", variable="count",
+                    type="integer", required=True,
+                ),
+                ToolArgument(
+                    flag="--mode", variable="mode",
+                    enum=["safe", "fast"], default="safe",
+                ),
+            ],
+        )
+        parser = create_argument_parser(tool)
+
+        args = parser.parse_args(["--project", "--count", "3", "--mode", "fast"])
+
+        assert args.project is True
+        assert args.count == 3
+        assert args.mode == "fast"
+
+    def test_parser_reuses_matching_legacy_boolean_universal_flag(self):
+        tool = Tool(
+            name="legacy",
+            arguments=[
+                ToolArgument(
+                    flag="--verbose", variable="verbose", type="bool"
+                )
+            ],
+        )
+
+        args = create_argument_parser(tool).parse_args(["--verbose"])
+
+        assert args.verbose is True
+        assert collect_custom_args(tool, args) == {"verbose": True}
+
+    def test_parser_rejects_incompatible_universal_flag_collision(self):
+        tool = Tool(
+            name="conflict",
+            arguments=[
+                ToolArgument(flag="--provider", variable="destination")
+            ],
+        )
+
+        with pytest.raises(ValueError, match="universal run options"):
+            create_argument_parser(tool)
+
     def test_parser_input_output_flags(self):
         tool = Tool(name="test")
         parser = create_argument_parser(tool)