fix(tests): fund validate_flow.py criteria at its 180s budget, guard it

tests/tasks/uipath-human-in-the-loop/e2e_01_invoice_approval_greenfield.yaml

@@ -103,7 +103,7 @@ success_criteria:
   - type: run_command
     description: "uip flow validate passes against the authored .flow file"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-human-in-the-loop/e2e_03_gdpr_compliance_greenfield.yaml

@@ -73,7 +73,7 @@ success_criteria:
   - type: run_command
     description: "uip flow validate passes against the authored .flow file"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-human-in-the-loop/e2e_06_invoice_approval_greenfield_simple.yaml

@@ -71,7 +71,7 @@ success_criteria:
   - type: run_command
     description: "uip flow validate passes against the authored .flow file"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-human-in-the-loop/quality_04_all_handles.yaml

@@ -54,7 +54,7 @@ success_criteria:
   - type: run_command
     description: "uip flow validate passes"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-human-in-the-loop/quality_05_priority_and_timeout.yaml

@@ -49,7 +49,7 @@ success_criteria:
   - type: run_command
     description: "uip flow validate passes"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-human-in-the-loop/quality_07_runtime_vars.yaml

@@ -45,7 +45,7 @@ success_criteria:
   - type: run_command
     description: "uip flow validate passes"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-human-in-the-loop/quality_08_variable_binding_fieldid.yaml

@@ -40,7 +40,7 @@ success_criteria:
   - type: run_command
     description: "uip flow validate passes"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-human-in-the-loop/quality_09_dev_mistake_wrong_type.yaml

@@ -38,7 +38,7 @@ success_criteria:
   - type: run_command
     description: "uip flow validate passes"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-human-in-the-loop/quality_10_dev_mistake_binding_direction.yaml

@@ -47,7 +47,7 @@ success_criteria:
   - type: run_command
     description: "uip flow validate passes"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-human-in-the-loop/quality_11_inout_field_access.yaml

@@ -39,7 +39,7 @@ success_criteria:
   - type: run_command
     description: "uip flow validate passes"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/_shared/test_criterion_budgets.py

@@ -32,6 +32,7 @@
 sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
 
 import flow_check  # noqa: E402
+import validate_flow  # noqa: E402
 
 _SUITE_ROOT = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
 # Split on ANY criterion so a block ends at its neighbour, and stop at the next
@@ -1369,3 +1370,69 @@ def test_while_condition_runs_once_more_than_the_body(tmp_path):
     it. Charging N left the criterion short by a whole debug_budget."""
     src = f"def main():\n    while {_ONE}:\n        {_ONE}\n"
     assert _price(src, tmp=str(tmp_path)) == 485 * 2 + 485 * 1
+
+
+# ── validate_flow.py budget ─────────────────────────────────────────────────
+#
+# `flow validate` refetches the tenant node manifest on every call: 8-14s
+# typical, 60-67s on the tail. validate_flow.py budgets itself at
+# `_DEFAULT_BUDGET_SECONDS` because the harness exports no criterion timeout,
+# so a criterion granting less is SIGKILLed on a tail hit (exit -1, empty
+# stdout) before the script can retry or name the file that stalled.
+#
+#     criterion timeout >= validate_flow._DEFAULT_BUDGET_SECONDS
+#
+# Every suite that runs the script is held to it. uipath-human-in-the-loop
+# reaches it through `reference: {directory: ../uipath-maestro-flow}`.
+
+_TASKS_ROOT = os.path.dirname(_SUITE_ROOT)
+_VALIDATE_SCRIPT = os.path.join(_SUITE_ROOT, "_shared", "validate_flow.py")
+
+
+def _validate_criteria():
+    """(yaml, timeout) per run_command criterion, in any suite, whose script
+    resolves to validate_flow.py."""
+    for root, _dirs, files in os.walk(_TASKS_ROOT):
+        for name in files:
+            if not name.endswith((".yaml", ".yml")):
+                continue
+            path = os.path.join(root, name)
+            text = open(path).read()
+            for block in _criterion_blocks(text, "run_command"):
+                command = _COMMAND.search(block)
+                if not command:
+                    continue
+                resolved = _script_path(root, command.group(1), text)
+                if resolved and os.path.samefile(resolved, _VALIDATE_SCRIPT):
+                    yield os.path.relpath(path, _TASKS_ROOT), _criterion_seconds(block)
+
+
+_VALIDATE_CASES = sorted(_validate_criteria())
+
+
+def test_every_validate_criterion_is_discovered():
+    """Counted a second way so a parser regression cannot quietly shrink the set."""
+    mentions = 0
+    for root, _dirs, files in os.walk(_TASKS_ROOT):
+        for name in files:
+            if not name.endswith((".yaml", ".yml")):
+                continue
+            with open(os.path.join(root, name)) as f:
+                mentions += sum(1 for ln in f if "command:" in ln and "validate_flow.py" in ln)
+    assert len(_VALIDATE_CASES) == mentions, (
+        f"{mentions - len(_VALIDATE_CASES)} validate_flow.py criteria went undiscovered"
+    )
+    assert len(_VALIDATE_CASES) > 40, f"only {len(_VALIDATE_CASES)} validate_flow.py criteria found"
+
+
+@pytest.mark.parametrize(
+    "yaml_path,criterion", _VALIDATE_CASES, ids=[y for y, _t in _VALIDATE_CASES]
+)
+def test_validate_criterion_funds_the_manifest_budget(yaml_path, criterion):
+    required = validate_flow._DEFAULT_BUDGET_SECONDS
+    assert criterion >= required, (
+        f"{yaml_path} grants {criterion}s to validate_flow.py, which budgets itself at "
+        f"{required}s. `flow validate` refetches the tenant manifest on every call "
+        "(8-14s typical, 60-67s tail), so a tail hit is SIGKILLed with exit -1 and an "
+        f"empty stdout. Raise the criterion to {required}s."
+    )

tests/tasks/uipath-maestro-flow/connector_features/complex_array.yaml

@@ -64,7 +64,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/dtl_load_by_default_false.yaml

@@ -54,7 +54,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/dtl_load_by_default_true.yaml

@@ -63,7 +63,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/enum.yaml

@@ -82,7 +82,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/generate_schema.yaml

@@ -56,7 +56,7 @@ success_criteria:
   - type: run_command
     description: "Persisted Flow artifact passes CLI validation"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 60
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/jdbc_databricks_query/jdbc_databricks_query.yaml

@@ -60,7 +60,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/multiselect.yaml

@@ -61,7 +61,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/non-catalog-http-fallback/non_catalog_http_fallback.yaml

@@ -48,7 +48,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/path_params.yaml

@@ -84,7 +84,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/query_params.yaml

@@ -58,7 +58,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/searchable_joins.yaml

@@ -54,7 +54,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/slack-http-fallback/slack_http_fallback.yaml

@@ -46,7 +46,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/testmanager_attachments/testmanager_attachments.yaml

@@ -66,7 +66,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 1.5
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/testmanager_crud_grounded/testmanager_crud_grounded.yaml

@@ -66,7 +66,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 1.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/testmanager_execution_results/testmanager_execution_results.yaml

@@ -71,7 +71,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 1.5
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/testmanager_generic_records/testmanager_generic_records.yaml

@@ -67,7 +67,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 1.5
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/testmanager_requirement_lifecycle/testmanager_requirement_lifecycle.yaml

@@ -68,7 +68,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 1.5
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/testmanager_testcase_lifecycle/testmanager_testcase_lifecycle.yaml

@@ -67,7 +67,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 1.5
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_features/testmanager_testset_lifecycle/testmanager_testset_lifecycle.yaml

@@ -69,7 +69,7 @@ success_criteria:
   - type: run_command
     description: "The completed flow validates"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 1.5
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/connector_trigger/webhook_waitfor_parallel.yaml

@@ -91,7 +91,7 @@ success_criteria:
   - type: run_command
     description: "Flow validates successfully"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 1.5
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/conversational/conversational_chat_loop.yaml

@@ -60,7 +60,7 @@ success_criteria:
   - type: run_command
     description: "uip maestro flow validate passes on the generated flow"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 60
+    timeout: 180
     expected_exit_code: 0
     weight: 3.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/e2e/jira_create_issue/jira_create_issue.yaml

@@ -66,7 +66,7 @@ success_criteria:
   - type: run_command
     description: "Flow validates successfully"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/e2e/jira_get_issue/jira_get_issue.yaml

@@ -66,7 +66,7 @@ success_criteria:
   - type: run_command
     description: "Flow validates successfully"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/e2e/jira_lifecycle/jira_lifecycle.yaml

@@ -79,7 +79,7 @@ success_criteria:
   - type: run_command
     description: "Flow validates successfully"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0

tests/tasks/uipath-maestro-flow/e2e/jira_search_triage/jira_search_triage.yaml

@@ -66,7 +66,7 @@ success_criteria:
   - type: run_command
     description: "Flow validates successfully"
     command: "python3 $REFERENCE_DIR/_shared/validate_flow.py"
-    timeout: 30
+    timeout: 180
     expected_exit_code: 0
     weight: 2.0
     pass_threshold: 1.0