fix(workflow): support integration: auto to follow project's initialized AI

src/specify_cli/__init__.py

@@ -1931,8 +1931,8 @@ def _read_integration_json(project_root: Path) -> dict[str, Any]:
         return {}
     try:
         data = json.loads(path.read_text(encoding="utf-8"))
-    except json.JSONDecodeError as exc:
-        console.print(f"[red]Error:[/red] {path} contains invalid JSON.")
+    except (json.JSONDecodeError, UnicodeDecodeError) as exc:
+        console.print(f"[red]Error:[/red] {path} contains invalid JSON or is not valid UTF-8.")
         console.print(f"Please fix or delete {INTEGRATION_JSON} and retry.")
         console.print(f"[dim]Details:[/dim] {exc}")
         raise typer.Exit(1)

src/specify_cli/workflows/engine.py

@@ -19,6 +19,12 @@
 
 import yaml
 
+from ..integration_state import (
+    INTEGRATION_JSON,
+    INTEGRATION_STATE_SCHEMA,
+    default_integration_key,
+    normalize_integration_state,
+)
 from .base import RunStatus, StepContext, StepResult, StepStatus
 
 
@@ -143,6 +149,35 @@ def validate_workflow(definition: WorkflowDefinition) -> list[str]:
                     f"Must be 'string', 'number', or 'boolean'."
                 )
 
+            # Validate the default eagerly so authoring mistakes (e.g. a
+            # default not in the declared enum, or a non-numeric default for
+            # a number input) surface at install/validation time instead of
+            # at workflow-execution time. ``"auto"`` for the integration
+            # input is a runtime-resolved sentinel, so only the
+            # enum-membership check is exempted for that exact case — the
+            # declared type is still enforced (e.g. ``type: number`` paired
+            # with ``default: "auto"`` is still rejected).
+            if "default" in input_def:
+                default_value = input_def["default"]
+                is_auto_integration = (
+                    input_name == "integration" and default_value == "auto"
+                )
+                validation_input_def: dict[str, Any] = input_def
+                if is_auto_integration and "enum" in input_def:
+                    validation_input_def = {
+                        key: value
+                        for key, value in input_def.items()
+                        if key != "enum"
+                    }
+                try:
+                    WorkflowEngine._coerce_input(
+                        input_name, default_value, validation_input_def
+                    )
+                except ValueError as exc:
+                    errors.append(
+                        f"Input {input_name!r} has invalid default: {exc}"
+                    )
+
     # -- Steps ------------------------------------------------------------
     if not isinstance(definition.steps, list):
         errors.append("'steps' must be a list.")
@@ -715,12 +750,74 @@ def _resolve_inputs(
                     name, provided[name], input_def
                 )
             elif "default" in input_def:
-                resolved[name] = input_def["default"]
+                default_value = self._resolve_default(name, input_def["default"])
+                # If the integration default could not be resolved against
+                # project state and falls back to the literal ``"auto"``
+                # sentinel, exempt it from enum-membership coercion so a
+                # workflow that lists specific integrations in ``enum`` does
+                # not crash at runtime — declared type is still enforced.
+                coerce_input_def = input_def
+                if (
+                    name == "integration"
+                    and default_value == "auto"
+                    and "enum" in input_def
+                ):
+                    coerce_input_def = {
+                        key: value
+                        for key, value in input_def.items()
+                        if key != "enum"
+                    }
+                resolved[name] = self._coerce_input(
+                    name, default_value, coerce_input_def
+                )
             elif input_def.get("required", False):
                 msg = f"Required input {name!r} not provided."
                 raise ValueError(msg)
         return resolved
 
+    def _resolve_default(self, name: str, default: Any) -> Any:
+        """Resolve special default sentinels against project state.
+
+        For the ``integration`` input, ``"auto"`` resolves to the integration
+        recorded in ``.specify/integration.json`` so workflows dispatch to the
+        AI the project was actually initialized with, instead of a hardcoded
+        value baked into the workflow YAML.
+        """
+        if name == "integration" and default == "auto":
+            resolved = self._load_project_integration()
+            if resolved is not None:
+                return resolved
+        return default
+
+    def _load_project_integration(self) -> str | None:
+        """Read the default integration key from ``.specify/integration.json``.
+
+        Honors the same schema guard as ``_read_integration_json`` (rejects
+        files whose ``integration_state_schema`` is newer than this CLI
+        supports) and reads the canonical normalized state, so modern
+        installs that record ``default_integration`` / ``installed_integrations``
+        resolve correctly under ``integration: auto``. Returns None when the
+        file is missing, malformed, or written by a newer CLI; callers are
+        expected to fall back to a literal default.
+        """
+        path = self.project_root / INTEGRATION_JSON
+        if not path.is_file():
+            return None
+        try:
+            data = json.loads(path.read_text(encoding="utf-8"))
+        except (json.JSONDecodeError, OSError, UnicodeDecodeError):
+            return None
+        if not isinstance(data, dict):
+            return None
+        schema = data.get("integration_state_schema")
+        if (
+            isinstance(schema, int)
+            and not isinstance(schema, bool)
+            and schema > INTEGRATION_STATE_SCHEMA
+        ):
+            return None
+        return default_integration_key(normalize_integration_state(data))
+
     @staticmethod
     def _coerce_input(
         name: str, value: Any, input_def: dict[str, Any]

tests/integrations/test_cli.py

@@ -1191,6 +1191,30 @@ def fail_search(self, **kwargs):
         assert "contains invalid JSON" in normalized_output
         assert "integration.json" in normalized_output
 
+    def test_search_rejects_non_utf8_integration_json_before_catalog_lookup(
+        self, tmp_path, monkeypatch
+    ):
+        """A non-UTF8 ``integration.json`` must surface a clear error and
+        avoid falling through to the catalog lookup, mirroring the malformed-JSON
+        case but for the ``UnicodeDecodeError`` branch in ``_read_integration_json``."""
+        project = self._make_project(tmp_path)
+        # 0xFF is invalid as the leading byte of any UTF-8 sequence, so
+        # ``Path.read_text(encoding="utf-8")`` raises ``UnicodeDecodeError``.
+        (project / ".specify" / "integration.json").write_bytes(b"\xff\xfe\x00\x00")
+
+        from specify_cli.integrations.catalog import IntegrationCatalog
+
+        def fail_search(self, **kwargs):
+            raise AssertionError("catalog search should not be called")
+
+        monkeypatch.setattr(IntegrationCatalog, "search", fail_search)
+
+        result = self._invoke(["integration", "search"], project)
+        normalized_output = _normalize_cli_output(result.output)
+        assert result.exit_code == 1
+        assert "not valid UTF-8" in normalized_output
+        assert "integration.json" in normalized_output
+
     def test_search_filters_by_tag(self, tmp_path, monkeypatch):
         project = self._make_project(tmp_path)
         self._patch_catalog(monkeypatch)