fix(workflows): reject a non-string 'integration'/'model' in command & prompt steps

src/specify_cli/workflows/steps/command/__init__.py

@@ -66,16 +66,52 @@ def execute(self, config: dict[str, Any], context: StepContext) -> StepResult:
         for key, value in input_data.items():
             resolved_input[key] = evaluate_expression(value, context)
 
-        # Resolve integration (step → workflow default → project default)
-        integration = config.get("integration") or context.default_integration
+        # Resolve integration (step → workflow default → project default).
+        # Fall back to the workflow default ONLY for a genuinely-unset value
+        # (missing / YAML-null / empty string). A ``config.get(...) or ...``
+        # would also swallow a falsey *non-string* ([], {}, 0, False), coercing
+        # it to the default before the guard below runs — so on an unvalidated
+        # execute() such a step would silently dispatch with the configured
+        # default instead of failing. Fall through instead, so every non-string
+        # reaches the type guard.
+        integration = config.get("integration")
+        if integration is None or integration == "":
+            integration = context.default_integration
         if integration and isinstance(integration, str) and "{{" in integration:
             integration = evaluate_expression(integration, context)
 
-        # Resolve model
-        model = config.get("model") or context.default_model
+        # Resolve model (same fallback rationale as 'integration' above).
+        model = config.get("model")
+        if model is None or model == "":
+            model = context.default_model
         if model and isinstance(model, str) and "{{" in model:
             model = evaluate_expression(model, context)
 
+        # A non-string integration/model — a literal list/dict/number that
+        # skipped validation, an unvalidated workflow-level default, or an
+        # expression that resolved to one — crashes downstream: get_integration()
+        # uses the value as a dict key (raw TypeError on an unhashable list/dict,
+        # even on a *validated* run) and build_exec_args() feeds model into the
+        # CLI argv. Fail the step with the contract error rather than taking down
+        # the whole run, mirroring the 'input'/'options' guards above. ``None``
+        # stays valid — it means "unset" and falls back to dispatch-not-possible.
+        if integration is not None and not isinstance(integration, str):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Command step {config.get('id', '?')!r}: 'integration' must "
+                    f"be a string, got {type(integration).__name__}."
+                ),
+            )
+        if model is not None and not isinstance(model, str):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Command step {config.get('id', '?')!r}: 'model' must be a "
+                    f"string, got {type(model).__name__}."
+                ),
+            )
+
         # Merge options (workflow defaults ← step overrides)
         options = dict(context.default_options)
         step_options = config.get("options", {})
@@ -217,4 +253,23 @@ def validate(self, config: dict[str, Any]) -> list[str]:
             errors.append(
                 f"Command step {config.get('id', '?')!r}: 'options' must be a mapping."
             )
+        # execute() passes 'integration' to get_integration(), which uses it as a
+        # dict key — a non-string (list/dict) raises a raw TypeError (unhashable),
+        # even on a validated run — and feeds 'model' into the CLI argv. Reject a
+        # literal non-string here, mirroring the sibling type checks. ``None``
+        # (an explicit ``integration:``/``model:`` YAML null) means "inherit the
+        # workflow default" and stays valid; an expression like "{{ ... }}" is
+        # still a str, so it stays valid too.
+        integration = config.get("integration")
+        if integration is not None and not isinstance(integration, str):
+            errors.append(
+                f"Command step {config.get('id', '?')!r}: 'integration' must be a "
+                f"string, got {type(integration).__name__}."
+            )
+        model = config.get("model")
+        if model is not None and not isinstance(model, str):
+            errors.append(
+                f"Command step {config.get('id', '?')!r}: 'model' must be a "
+                f"string, got {type(model).__name__}."
+            )
         return errors

src/specify_cli/workflows/steps/prompt/__init__.py

@@ -42,16 +42,52 @@ def execute(self, config: dict[str, Any], context: StepContext) -> StepResult:
         if not isinstance(prompt, str):
             prompt = str(prompt)
 
-        # Resolve integration (step → workflow default)
-        integration = config.get("integration") or context.default_integration
+        # Resolve integration (step → workflow default).
+        # Fall back to the workflow default ONLY for a genuinely-unset value
+        # (missing / YAML-null / empty string). A ``config.get(...) or ...``
+        # would also swallow a falsey *non-string* ([], {}, 0, False), coercing
+        # it to the default before the guard below runs — so on an unvalidated
+        # execute() such a step would silently dispatch with the configured
+        # default instead of failing. Fall through instead, so every non-string
+        # reaches the type guard.
+        integration = config.get("integration")
+        if integration is None or integration == "":
+            integration = context.default_integration
         if integration and isinstance(integration, str) and "{{" in integration:
             integration = evaluate_expression(integration, context)
 
-        # Resolve model
-        model = config.get("model") or context.default_model
+        # Resolve model (same fallback rationale as 'integration' above).
+        model = config.get("model")
+        if model is None or model == "":
+            model = context.default_model
         if model and isinstance(model, str) and "{{" in model:
             model = evaluate_expression(model, context)
 
+        # A non-string integration/model — a literal list/dict/number that
+        # skipped validation, an unvalidated workflow-level default, or an
+        # expression that resolved to one — crashes downstream: get_integration()
+        # uses the value as a dict key (raw TypeError on an unhashable list/dict,
+        # even on a *validated* run) and build_exec_args() feeds model into the
+        # CLI argv. Fail the step with the contract error rather than taking down
+        # the whole run. ``None`` stays valid — it means "unset" and falls back
+        # to dispatch-not-possible.
+        if integration is not None and not isinstance(integration, str):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Prompt step {config.get('id', '?')!r}: 'integration' must "
+                    f"be a string, got {type(integration).__name__}."
+                ),
+            )
+        if model is not None and not isinstance(model, str):
+            return StepResult(
+                status=StepStatus.FAILED,
+                error=(
+                    f"Prompt step {config.get('id', '?')!r}: 'model' must be a "
+                    f"string, got {type(model).__name__}."
+                ),
+            )
+
         # Attempt CLI dispatch
         dispatch_result = self._try_dispatch(
             prompt, integration, model, context
@@ -172,4 +208,23 @@ def validate(self, config: dict[str, Any]) -> list[str]:
                 f"Prompt step {config.get('id', '?')!r}: 'prompt' must be a "
                 f"string, got {type(config['prompt']).__name__}."
             )
+        # execute() passes 'integration' to get_integration(), which uses it as a
+        # dict key — a non-string (list/dict) raises a raw TypeError (unhashable),
+        # even on a validated run — and feeds 'model' into the CLI argv. Reject a
+        # literal non-string here, mirroring the 'prompt' check above. ``None``
+        # (an explicit ``integration:``/``model:`` YAML null) means "inherit the
+        # workflow default" and stays valid; an expression like "{{ ... }}" is
+        # still a str, so it stays valid too.
+        integration = config.get("integration")
+        if integration is not None and not isinstance(integration, str):
+            errors.append(
+                f"Prompt step {config.get('id', '?')!r}: 'integration' must be a "
+                f"string, got {type(integration).__name__}."
+            )
+        model = config.get("model")
+        if model is not None and not isinstance(model, str):
+            errors.append(
+                f"Prompt step {config.get('id', '?')!r}: 'model' must be a "
+                f"string, got {type(model).__name__}."
+            )
         return errors

tests/test_workflows.py

@@ -1026,6 +1026,41 @@ def test_validate_rejects_non_mapping_input_and_options(self):
         assert res_opt.status is StepStatus.FAILED
         assert "'options' must be a mapping" in (res_opt.error or "")
 
+    @pytest.mark.parametrize("bad", [["claude"], {"a": 1}, 5, True])
+    def test_validate_rejects_non_string_integration_and_model(self, bad):
+        """A non-string 'integration'/'model' must be rejected at validation.
+
+        execute() passes 'integration' to get_integration(), which uses it as a
+        dict key — an unhashable list/dict raises a raw TypeError there, even on
+        a validated run — and feeds 'model' into the CLI argv. Mirrors the
+        'command'/'input'/'options' type checks.
+        """
+        from specify_cli.workflows.steps.command import CommandStep
+
+        step = CommandStep()
+        errs = step.validate({"id": "c", "command": "/x", "integration": bad})
+        assert any("'integration' must be a string" in e for e in errs), bad
+        errs = step.validate({"id": "c", "command": "/x", "model": bad})
+        assert any("'model' must be a string" in e for e in errs), bad
+
+    def test_validate_accepts_none_and_expression_integration_model(self):
+        """An explicit YAML-null (inherit default) or a '{{ ... }}' expression
+        integration/model stays valid — only literal non-strings are rejected."""
+        from specify_cli.workflows.steps.command import CommandStep
+
+        step = CommandStep()
+        assert step.validate(
+            {"id": "c", "command": "/x", "integration": None, "model": None}
+        ) == []
+        assert step.validate(
+            {
+                "id": "c",
+                "command": "/x",
+                "integration": "{{ inputs.agent }}",
+                "model": "{{ inputs.model }}",
+            }
+        ) == []
+
     def test_validate_rejects_non_string_command(self):
         from specify_cli.workflows.steps.command import CommandStep
 
@@ -1061,6 +1096,55 @@ def test_execute_non_string_command_fails_cleanly(self):
                 assert result.status is StepStatus.FAILED, bad
                 assert "'command' must be a string" in (result.error or ""), bad
 
+    def test_execute_non_string_integration_fails_loudly(self):
+        """On an unvalidated run, an unhashable 'integration' would crash
+        get_integration() (dict.get on a list) with a raw TypeError. execute()
+        must fail the step with the contract error instead."""
+        from specify_cli.workflows.steps.command import CommandStep
+        from specify_cli.workflows.base import StepContext, StepStatus
+
+        step = CommandStep()
+        res = step.execute(
+            {"id": "c", "command": "speckit.specify", "integration": ["claude"]},
+            StepContext(),
+        )
+        assert res.status is StepStatus.FAILED
+        assert "'integration' must be a string" in (res.error or "")
+        # non-string model likewise fails before build_exec_args
+        res = step.execute(
+            {"id": "c", "command": "speckit.specify", "integration": "claude", "model": ["m"]},
+            StepContext(),
+        )
+        assert res.status is StepStatus.FAILED
+        assert "'model' must be a string" in (res.error or "")
+
+    @pytest.mark.parametrize("falsey", [[], {}, 0, False])
+    def test_execute_falsey_non_string_integration_fails_loudly(self, falsey):
+        """A *falsey* non-string ([], {}, 0, False) must fail the step, not be
+        swallowed by an ``or``-fallback to the workflow default.
+
+        A ``config.get('integration') or context.default_integration`` coerces a
+        falsey non-string to the default *before* the type guard runs, so with a
+        configured default the step would silently dispatch using the wrong
+        integration instead of surfacing the contract error. The default is set
+        here so a regression dispatches rather than fails-not-possible."""
+        from specify_cli.workflows.steps.command import CommandStep
+        from specify_cli.workflows.base import StepContext, StepStatus
+
+        step = CommandStep()
+        ctx = StepContext(default_integration="claude", default_model="sonnet")
+        res = step.execute(
+            {"id": "c", "command": "speckit.specify", "integration": falsey}, ctx
+        )
+        assert res.status is StepStatus.FAILED, falsey
+        assert "'integration' must be a string" in (res.error or ""), falsey
+        # a falsey non-string model likewise reaches the guard
+        res = step.execute(
+            {"id": "c", "command": "speckit.specify", "model": falsey}, ctx
+        )
+        assert res.status is StepStatus.FAILED, falsey
+        assert "'model' must be a string" in (res.error or ""), falsey
+
     def test_step_override_integration(self):
         from unittest.mock import patch
         from specify_cli.workflows.steps.command import CommandStep
@@ -1448,6 +1532,82 @@ def test_validate_accepts_expression_prompt(self):
         )
         assert errors == []
 
+    @pytest.mark.parametrize("bad", [["claude"], {"a": 1}, 5, True])
+    def test_validate_rejects_non_string_integration_and_model(self, bad):
+        """A non-string 'integration'/'model' must be rejected at validation.
+
+        execute() passes 'integration' to get_integration(), which uses it as a
+        dict key — an unhashable list/dict raises a raw TypeError there, even on
+        a validated run — and feeds 'model' into the CLI argv."""
+        from specify_cli.workflows.steps.prompt import PromptStep
+
+        step = PromptStep()
+        errs = step.validate({"id": "p", "prompt": "hi", "integration": bad})
+        assert any("'integration' must be a string" in e for e in errs), bad
+        errs = step.validate({"id": "p", "prompt": "hi", "model": bad})
+        assert any("'model' must be a string" in e for e in errs), bad
+
+    def test_validate_accepts_none_and_expression_integration_model(self):
+        """An explicit YAML-null (inherit default) or a '{{ ... }}' expression
+        integration/model stays valid — only literal non-strings are rejected."""
+        from specify_cli.workflows.steps.prompt import PromptStep
+
+        step = PromptStep()
+        assert step.validate(
+            {"id": "p", "prompt": "hi", "integration": None, "model": None}
+        ) == []
+        assert step.validate(
+            {
+                "id": "p",
+                "prompt": "hi",
+                "integration": "{{ inputs.agent }}",
+                "model": "{{ inputs.model }}",
+            }
+        ) == []
+
+    def test_execute_non_string_integration_fails_loudly(self):
+        """On an unvalidated run, an unhashable 'integration' would crash
+        get_integration() (dict.get on a dict) with a raw TypeError. execute()
+        must fail the step with the contract error instead."""
+        from specify_cli.workflows.steps.prompt import PromptStep
+        from specify_cli.workflows.base import StepContext, StepStatus
+
+        step = PromptStep()
+        res = step.execute(
+            {"id": "p", "prompt": "hi", "integration": {"a": 1}}, StepContext()
+        )
+        assert res.status is StepStatus.FAILED
+        assert "'integration' must be a string" in (res.error or "")
+        res = step.execute(
+            {"id": "p", "prompt": "hi", "integration": "claude", "model": ["m"]},
+            StepContext(),
+        )
+        assert res.status is StepStatus.FAILED
+        assert "'model' must be a string" in (res.error or "")
+
+    @pytest.mark.parametrize("falsey", [[], {}, 0, False])
+    def test_execute_falsey_non_string_integration_fails_loudly(self, falsey):
+        """A *falsey* non-string ([], {}, 0, False) must fail the step, not be
+        swallowed by an ``or``-fallback to the workflow default.
+
+        A ``config.get('integration') or context.default_integration`` coerces a
+        falsey non-string to the default *before* the type guard runs, so with a
+        configured default the step would silently dispatch using the wrong
+        integration instead of surfacing the contract error. The default is set
+        here so a regression dispatches rather than fails-not-possible."""
+        from specify_cli.workflows.steps.prompt import PromptStep
+        from specify_cli.workflows.base import StepContext, StepStatus
+
+        step = PromptStep()
+        ctx = StepContext(default_integration="claude", default_model="sonnet")
+        res = step.execute({"id": "p", "prompt": "hi", "integration": falsey}, ctx)
+        assert res.status is StepStatus.FAILED, falsey
+        assert "'integration' must be a string" in (res.error or ""), falsey
+        # a falsey non-string model likewise reaches the guard
+        res = step.execute({"id": "p", "prompt": "hi", "model": falsey}, ctx)
+        assert res.status is StepStatus.FAILED, falsey
+        assert "'model' must be a string" in (res.error or ""), falsey
+
 
 class TestShellStep:
     """Test the shell step type."""