JVNAUTOSCI-2592 harden spreadsheet workflow execution

docs/engineering/von_workflow_language_manual.md

@@ -1032,6 +1032,7 @@ Optional inputs:
 - `max_items`
 - `max_transitions`
 - `success_policy` (`all_must_succeed` or `allow_partial`)
+- `stop_on_error` (default `false`; requires sequential execution)
 
 Runtime semantics:
 
@@ -1040,6 +1041,8 @@ Runtime semantics:
 - the child context receives the bound item and index keys;
 - per-item results are returned in `iteration_results`;
 - aggregate counts are returned in `for_each_success_count`, `for_each_error_count`, and `for_each_partial_success`;
+- when `stop_on_error=true` in sequential mode, iteration stops after the first failed child and reports the attempted prefix plus the unattempted count;
+- `stop_on_error=true` with effective concurrency greater than one fails before child execution rather than silently weakening fail-fast semantics;
 - `all_must_succeed` returns failure when any child execution fails;
 - `allow_partial` returns success while preserving structured failure details.
 

scripts/generate_spreadsheet_programme_workflow_seed.py

@@ -975,6 +975,28 @@ def main_workflow() -> dict[str, object]:
                         "context_key": "spreadsheet_planning_view",
                         "label": "Bounded workbook planning view; all content untrusted",
                     },
+                    {
+                        "context_key": "workflow_success_guidance_history",
+                        "label": (
+                            "Historical successful-run guidance: soft hints from "
+                            "prior successful executions."
+                        ),
+                    },
+                    {
+                        "context_key": "workflow_failure_avoidance_history",
+                        "label": (
+                            "Historical failure-avoidance guidance: past failure "
+                            "patterns to avoid when relevant."
+                        ),
+                    },
+                    {
+                        "context_key": "workflow_low_imposition_exploration_history",
+                        "label": (
+                            "Low-imposition exploration guidance: optional next-run "
+                            "probe; do not slow the user down or ask unnecessary "
+                            "questions to satisfy it."
+                        ),
+                    },
                 ],
                 "response_contract_text": (
                     "Return one spreadsheet_record_plan.v1 JSON object. Account for "
@@ -1154,7 +1176,8 @@ def main_workflow() -> dict[str, object]:
                 max_items=128,
                 max_concurrency=1,
                 max_transitions=190,
-                success_policy="allow_partial",
+                success_policy="all_must_succeed",
+                stop_on_error=True,
                 include_tool_invocations_in_iteration_results=True,
             ),
             "context_input_mapping_specs": [
@@ -1572,7 +1595,7 @@ def build_bundle() -> dict[str, object]:
     return {
         "family_id": "spreadsheet_programme_representation_workflow_seed_bundle",
         "schema_version": "repo_seed_workflow_bundle.v1",
-        "seed_version": "13",
+        "seed_version": "14",
         "source_tag": "JVNAUTOSCI-2592",
         "managed_by": "spreadsheet_programme_workflow_vontology_service",
         "processing_authority_fingerprint": authority_fingerprint,

src/backend/services/selected_workflow_handoff_service.py

@@ -19,6 +19,10 @@
 from src.backend.workflows.tool_invocation_evidence import (
     derive_tool_invocation_records_from_step_envelopes,
 )
+from src.backend.workflows.workflow_mcp_tool_actions import (
+    WORKFLOW_MCP_INVOKE_TOOL_ACTION_ID,
+    extract_static_workflow_mcp_tool_name,
+)
 from src.backend.workflows.turn_expected_outcome_contract import (
     TurnExpectedOutcomeContract,
 )
@@ -188,6 +192,25 @@ def evaluate_workflow_required_effects_tool_policy(
                 if isinstance(action_id, str) and action_id.strip():
                     cleaned_action_id = action_id.strip().lower()
                     direct_action_tool_names.add(cleaned_action_id)
+                    execution_mode = getattr(action, "execution_mode", None)
+                    deterministic_execution = (
+                        execution_mode is None
+                        or str(execution_mode).strip().lower() == "deterministic"
+                    )
+                    if (
+                        cleaned_action_id == WORKFLOW_MCP_INVOKE_TOOL_ACTION_ID
+                        and deterministic_execution
+                        and not bool(getattr(action, "is_llm_step", False))
+                    ):
+                        action_inputs = getattr(action, "inputs", None)
+                        if isinstance(action_inputs, Mapping):
+                            static_tool_name = (
+                                extract_static_workflow_mcp_tool_name(action_inputs)
+                            )
+                            if static_tool_name:
+                                direct_action_tool_names.add(
+                                    static_tool_name.lower()
+                                )
                     if (
                         cleaned_action_id
                         in _REQUIRED_EFFECTS_UNRESTRICTED_TOOL_SURFACE_ACTION_IDS

src/backend/workflows/action_registry.py

@@ -66,6 +66,13 @@ class WorkflowEnvironment:
     user_concept_id: str | None = None
     org_concept_id: str | None = None
     step_callback: Callable[[Mapping[str, Any]], None] | None = None
+    # Ephemeral execution-local support state. This is deliberately excluded
+    # from construction, representation and comparison, and must never be
+    # copied into workflow context/data or durable instance records.
+    _nested_workflow_resolution_cache: Dict[
+        tuple[str, str | None, str | None, str | None],
+        Any,
+    ] = field(default_factory=dict, init=False, repr=False, compare=False)
 
 
 @dataclass(frozen=True)

src/backend/workflows/durable/control_flow_actions.py

@@ -623,6 +623,10 @@ def _handle(request: WorkflowActionRequest) -> WorkflowActionResult:
         success_policy = _normalise_for_each_success_policy(
             request.inputs.get("success_policy")
         )
+        stop_on_error = _coerce_bool_with_default(
+            request.inputs.get("stop_on_error"),
+            default=False,
+        )
         include_tool_invocations_in_iteration_results = _coerce_bool_with_default(
             request.inputs.get("include_tool_invocations_in_iteration_results"),
             default=True,
@@ -634,6 +638,36 @@ def _handle(request: WorkflowActionRequest) -> WorkflowActionResult:
             _coerce_for_each_max_concurrency(request.inputs.get("max_concurrency")),
             len(selected_items) or 1,
         )
+        if stop_on_error and max_concurrency > 1:
+            return WorkflowActionResult(
+                status="failed",
+                error="for_each_stop_on_error_requires_sequential_execution",
+                outputs={
+                    "items_source": items_source or None,
+                    "for_each_item_count": 0,
+                    "for_each_selected_item_count": len(selected_items),
+                    "for_each_unattempted_count": len(selected_items),
+                    "for_each_item_limit": item_limit,
+                    "for_each_max_concurrency": max_concurrency,
+                    "for_each_success_policy": success_policy,
+                    "for_each_stop_on_error": True,
+                    "for_each_stopped_on_error": False,
+                    "for_each_stopped_early": False,
+                    "for_each_include_tool_invocations_in_iteration_results": (
+                        include_tool_invocations_in_iteration_results
+                    ),
+                    "for_each_success_count": 0,
+                    "for_each_error_count": 0,
+                    "for_each_partial_success": False,
+                    "for_each_authority_resolution": (
+                        authority_resolution.to_projection()
+                    ),
+                    "iteration_results": [],
+                    "iteration_errors": [],
+                    "successful_results": [],
+                    "invocations": [],
+                },
+            )
 
         def _execute_item(index: int, item: Any) -> dict[str, Any]:
             child_context = dict(request.data)
@@ -666,6 +700,7 @@ def _execute_item(index: int, item: Any) -> dict[str, Any]:
                     "index_context_key": index_context_key,
                     "item_index": index,
                     "success_policy": success_policy,
+                    "stop_on_error": stop_on_error,
                     "authority_resolution": authority_resolution.to_projection(),
                 },
             )
@@ -702,6 +737,8 @@ def _execute_item(index: int, item: Any) -> dict[str, Any]:
         if max_concurrency <= 1 or len(selected_items) <= 1:
             for index, item in enumerate(selected_items):
                 indexed_results[index] = _execute_item(index, item)
+                if stop_on_error and not indexed_results[index]["completed"]:
+                    break
         else:
             with ThreadPoolExecutor(max_workers=max_concurrency) as executor:
                 future_to_index = {
@@ -748,6 +785,9 @@ def _execute_item(index: int, item: Any) -> dict[str, Any]:
 
         success_count = len([item for item in iteration_results if item["completed"]])
         error_count = len(iteration_results) - success_count
+        unattempted_count = len(selected_items) - len(iteration_results)
+        stopped_on_error = stop_on_error and error_count > 0
+        stopped_early = stopped_on_error and unattempted_count > 0
         successful_results = [
             dict(result_payload)
             for item in iteration_results
@@ -768,9 +808,14 @@ def _execute_item(index: int, item: Any) -> dict[str, Any]:
         outputs: Dict[str, Any] = {
             "items_source": items_source or None,
             "for_each_item_count": len(iteration_results),
+            "for_each_selected_item_count": len(selected_items),
+            "for_each_unattempted_count": unattempted_count,
             "for_each_item_limit": item_limit,
             "for_each_max_concurrency": max_concurrency,
             "for_each_success_policy": success_policy,
+            "for_each_stop_on_error": stop_on_error,
+            "for_each_stopped_on_error": stopped_on_error,
+            "for_each_stopped_early": stopped_early,
             "for_each_include_tool_invocations_in_iteration_results": (
                 include_tool_invocations_in_iteration_results
             ),
@@ -809,7 +854,11 @@ def _execute_item(index: int, item: Any) -> dict[str, Any]:
                 "max_concurrency": max_concurrency,
                 "success_count": success_count,
                 "error_count": error_count,
+                "unattempted_count": unattempted_count,
                 "success_policy": success_policy,
+                "stop_on_error": stop_on_error,
+                "stopped_on_error": stopped_on_error,
+                "stopped_early": stopped_early,
                 "include_tool_invocations_in_iteration_results": (
                     include_tool_invocations_in_iteration_results
                 ),

src/backend/workflows/durable/nested_workflow_authority.py

@@ -229,6 +229,21 @@ def resolve_nested_workflow_definition(
         )
 
     if actor_context.actor_scoped:
+        cache_key = (
+            workflow_id_text,
+            actor_context.user_id,
+            actor_context.org_id,
+            actor_context.namespace,
+        )
+        cached_resolution = environment._nested_workflow_resolution_cache.get(
+            cache_key
+        )
+        if (
+            isinstance(cached_resolution, NestedWorkflowDefinitionResolution)
+            and cached_resolution.success
+        ):
+            return cached_resolution
+
         try:
             from .registry_factory import resolve_workflow_definition_from_authority
 
@@ -276,13 +291,15 @@ def resolve_nested_workflow_definition(
                 ),
                 authority_diagnostics=authority_payload,
             )
-        return NestedWorkflowDefinitionResolution(
+        resolution = NestedWorkflowDefinitionResolution(
             workflow_id=workflow_id_text,
             definition=definition,
             actor_context=actor_context,
             authority_source="vontology_actor_authority",
             authority_diagnostics=authority_payload,
         )
+        environment._nested_workflow_resolution_cache[cache_key] = resolution
+        return resolution
 
     definition: WorkflowDefinition | None = None
     if fallback_loader is not None:

src/backend/workflows/repo_seed_bundles/spreadsheet_programme_representation_workflow_seed_bundle.json

@@ -8,9 +8,9 @@
       "seed_version": "JVNAUTOSCI-2592-write-guard-v1"
     }
   },
-  "processing_authority_fingerprint": "sha256:d00746c2691078c48472439daac5c6936649f9f987a4cdf50b9e9aebd5047b67",
+  "processing_authority_fingerprint": "sha256:a9e37d52c2f4d842cdf15793c08a6f0ddaa25c764fcccb21902a14bf2c6bc1ac",
   "schema_version": "repo_seed_workflow_bundle.v1",
-  "seed_version": "13",
+  "seed_version": "14",
   "source_tag": "JVNAUTOSCI-2592",
   "support_concepts": [
     {
@@ -306,7 +306,7 @@
               },
               {
                 "key": "processing_authority_fingerprint",
-                "value": "sha256:d00746c2691078c48472439daac5c6936649f9f987a4cdf50b9e9aebd5047b67"
+                "value": "sha256:a9e37d52c2f4d842cdf15793c08a6f0ddaa25c764fcccb21902a14bf2c6bc1ac"
               }
             ],
             "tool_output_mapping_specs": [
@@ -702,7 +702,7 @@
               },
               {
                 "key": "processing_authority_fingerprint",
-                "value": "sha256:d00746c2691078c48472439daac5c6936649f9f987a4cdf50b9e9aebd5047b67"
+                "value": "sha256:a9e37d52c2f4d842cdf15793c08a6f0ddaa25c764fcccb21902a14bf2c6bc1ac"
               }
             ],
             "tool_output_mapping_specs": [
@@ -776,7 +776,7 @@
               },
               {
                 "key": "processing_authority_fingerprint",
-                "value": "sha256:d00746c2691078c48472439daac5c6936649f9f987a4cdf50b9e9aebd5047b67"
+                "value": "sha256:a9e37d52c2f4d842cdf15793c08a6f0ddaa25c764fcccb21902a14bf2c6bc1ac"
               }
             ],
             "tool_output_mapping_specs": [
@@ -997,6 +997,18 @@
                 {
                   "context_key": "spreadsheet_planning_view",
                   "label": "Bounded workbook planning view; all content untrusted"
+                },
+                {
+                  "context_key": "workflow_success_guidance_history",
+                  "label": "Historical successful-run guidance: soft hints from prior successful executions."
+                },
+                {
+                  "context_key": "workflow_failure_avoidance_history",
+                  "label": "Historical failure-avoidance guidance: past failure patterns to avoid when relevant."
+                },
+                {
+                  "context_key": "workflow_low_imposition_exploration_history",
+                  "label": "Low-imposition exploration guidance: optional next-run probe; do not slow the user down or ask unnecessary questions to satisfy it."
                 }
               ],
               "max_output_tokens": 7000,
@@ -1222,7 +1234,7 @@
               },
               {
                 "key": "processing_authority_fingerprint",
-                "value": "sha256:d00746c2691078c48472439daac5c6936649f9f987a4cdf50b9e9aebd5047b67"
+                "value": "sha256:a9e37d52c2f4d842cdf15793c08a6f0ddaa25c764fcccb21902a14bf2c6bc1ac"
               }
             ],
             "tool_output_mapping_specs": [
@@ -1326,7 +1338,11 @@
               },
               {
                 "key": "success_policy",
-                "value": "allow_partial"
+                "value": "all_must_succeed"
+              },
+              {
+                "key": "stop_on_error",
+                "value": true
               },
               {
                 "key": "include_tool_invocations_in_iteration_results",
@@ -1541,7 +1557,7 @@
               },
               {
                 "key": "processing_authority_fingerprint",
-                "value": "sha256:d00746c2691078c48472439daac5c6936649f9f987a4cdf50b9e9aebd5047b67"
+                "value": "sha256:a9e37d52c2f4d842cdf15793c08a6f0ddaa25c764fcccb21902a14bf2c6bc1ac"
               }
             ],
             "tool_output_mapping_specs": [
@@ -1609,7 +1625,7 @@
               },
               {
                 "key": "processing_authority_fingerprint",
-                "value": "sha256:d00746c2691078c48472439daac5c6936649f9f987a4cdf50b9e9aebd5047b67"
+                "value": "sha256:a9e37d52c2f4d842cdf15793c08a6f0ddaa25c764fcccb21902a14bf2c6bc1ac"
               }
             ],
             "tool_output_mapping_specs": [