diff --git a/platform/components/__init__.py b/platform/components/__init__.py
index f295dfb..f60a1d0 100644
--- a/platform/components/__init__.py
+++ b/platform/components/__init__.py
@@ -31,6 +31,7 @@ def get_component_factory(component_type: str):
 from components import (  # noqa: E402, F401
     agent,
     ai_model,
+    assertion,
     categorizer,
     code,
     control_flow,
diff --git a/platform/components/assertion.py b/platform/components/assertion.py
new file mode 100644
index 0000000..5c03b39
--- /dev/null
+++ b/platform/components/assertion.py
@@ -0,0 +1,43 @@
+"""Assertion component — evaluates all rules and routes pass/fail."""
+
+from __future__ import annotations
+
+from components import register
+from components.operators import evaluate_rules
+
+
+@register("assertion")
+def assertion_factory(node):
+    """Build an assertion graph node that checks all rules against state."""
+    extra = node.component_config.extra_config
+    rules = extra.get("rules", [])
+
+    def assertion_node(state: dict) -> dict:
+        if not rules:
+            return {
+                "_route": "pass",
+                "output": {"passed": True, "results": []},
+            }
+
+        results = evaluate_rules(rules, state, mode="all")
+
+        results_list = []
+        for rule, result in zip(rules, results):
+            results_list.append({
+                "check": f"{rule.get('field', '')} {rule.get('operator', 'equals')} {rule.get('value', '')}",
+                "passed": result["passed"],
+                "actual": result["actual_value"],
+                "expected": rule.get("value", ""),
+            })
+
+        all_passed = all(r["passed"] for r in results)
+
+        return {
+            "_route": "pass" if all_passed else "fail",
+            "output": {
+                "passed": all_passed,
+                "results": results_list,
+            },
+        }
+
+    return assertion_node
diff --git a/platform/components/operators.py b/platform/components/operators.py
index 42c4240..2868b86 100644
--- a/platform/components/operators.py
+++ b/platform/components/operators.py
@@ -101,3 +101,54 @@ def _resolve_field(path: str, state: dict):
         else:
             return None
     return current
+
+
+def evaluate_rules(
+    rules: list[dict],
+    state: dict,
+    mode: str = "first_match",
+) -> str | list[dict]:
+    """Evaluate a list of rules against workflow state.
+
+    Args:
+        rules: List of rule dicts with keys: id, field, operator, value.
+        state: The workflow state dict to evaluate against.
+        mode: "first_match" returns the first matching rule's id (str).
+              "all" evaluates every rule and returns a list of result dicts.
+
+    Returns:
+        mode="first_match": The matching rule's id (str), or "" if none match.
+        mode="all": List of {"rule_id": str, "passed": bool, "actual_value": any}.
+    """
+    if mode == "all":
+        results = []
+        for rule in rules:
+            field_path = rule.get("field", "")
+            operator = rule.get("operator", "equals")
+            value = rule.get("value", "")
+            rule_id = rule.get("id", "")
+
+            field_val = _resolve_field(field_path, state)
+            op_fn = OPERATORS.get(operator)
+            passed = bool(op_fn(field_val, value)) if op_fn else False
+
+            results.append({
+                "rule_id": rule_id,
+                "passed": passed,
+                "actual_value": field_val,
+            })
+        return results
+
+    # mode="first_match" (default)
+    for rule in rules:
+        field_path = rule.get("field", "")
+        operator = rule.get("operator", "equals")
+        value = rule.get("value", "")
+        rule_id = rule.get("id", "")
+
+        field_val = _resolve_field(field_path, state)
+        op_fn = OPERATORS.get(operator)
+        if op_fn and op_fn(field_val, value):
+            return rule_id
+
+    return ""
diff --git a/platform/components/switch.py b/platform/components/switch.py
index b087dbc..efa05bc 100644
--- a/platform/components/switch.py
+++ b/platform/components/switch.py
@@ -3,7 +3,7 @@
 from __future__ import annotations
 
 from components import register
-from components.operators import OPERATORS, UNARY_OPERATORS, _resolve_field
+from components.operators import _resolve_field, evaluate_rules
 
 
 @register("switch")
@@ -17,18 +17,7 @@ def switch_factory(node):
         enable_fallback = extra.get("enable_fallback", False)
 
         def switch_node(state: dict) -> dict:
-            route = ""
-            for rule in rules:
-                field_path = rule.get("field", "")
-                operator = rule.get("operator", "equals")
-                value = rule.get("value", "")
-                rule_id = rule.get("id", "")
-
-                field_val = _resolve_field(field_path, state)
-                op_fn = OPERATORS.get(operator)
-                if op_fn and op_fn(field_val, value):
-                    route = rule_id
-                    break
+            route = evaluate_rules(rules, state, mode="first_match")
 
             if not route and enable_fallback:
                 route = "__other__"
diff --git a/platform/frontend/src/features/workflows/components/NodeDetailsPanel.tsx b/platform/frontend/src/features/workflows/components/NodeDetailsPanel.tsx
index f33d54c..843790c 100644
--- a/platform/frontend/src/features/workflows/components/NodeDetailsPanel.tsx
+++ b/platform/frontend/src/features/workflows/components/NodeDetailsPanel.tsx
@@ -21,6 +21,7 @@ import CodeMirrorExpressionEditor from "@/components/CodeMirrorExpressionEditor"
 import PopoutWindow from "@/components/PopoutWindow"
 import type { CodeMirrorLanguage } from "@/components/CodeMirrorEditor"
 import type { WorkflowNode, WorkflowDetail, SwitchRule, FilterRule, ScheduleJobInfo } from "@/types/models"
+import RuleEditor from "./RuleEditor"
 
 interface Props {
   slug: string
@@ -31,52 +32,6 @@ interface Props {
 
 const TRIGGER_TYPES = ["trigger_telegram", "trigger_schedule", "trigger_manual", "trigger_workflow", "trigger_error", "trigger_chat"]
 
-const OPERATOR_OPTIONS = [
-  { group: "Universal", options: [
-    { value: "exists", label: "Exists" },
-    { value: "does_not_exist", label: "Does not exist" },
-    { value: "is_empty", label: "Is empty" },
-    { value: "is_not_empty", label: "Is not empty" },
-  ]},
-  { group: "String", options: [
-    { value: "equals", label: "Equals" },
-    { value: "not_equals", label: "Not equals" },
-    { value: "contains", label: "Contains" },
-    { value: "not_contains", label: "Not contains" },
-    { value: "starts_with", label: "Starts with" },
-    { value: "not_starts_with", label: "Not starts with" },
-    { value: "ends_with", label: "Ends with" },
-    { value: "not_ends_with", label: "Not ends with" },
-    { value: "matches_regex", label: "Matches regex" },
-    { value: "not_matches_regex", label: "Not matches regex" },
-  ]},
-  { group: "Number", options: [
-    { value: "gt", label: "Greater than" },
-    { value: "lt", label: "Less than" },
-    { value: "gte", label: "Greater or equal" },
-    { value: "lte", label: "Less or equal" },
-  ]},
-  { group: "Datetime", options: [
-    { value: "after", label: "After" },
-    { value: "before", label: "Before" },
-    { value: "after_or_equal", label: "After or equal" },
-    { value: "before_or_equal", label: "Before or equal" },
-  ]},
-  { group: "Boolean", options: [
-    { value: "is_true", label: "Is true" },
-    { value: "is_false", label: "Is false" },
-  ]},
-  { group: "Array", options: [
-    { value: "length_eq", label: "Length equals" },
-    { value: "length_neq", label: "Length not equals" },
-    { value: "length_gt", label: "Length greater than" },
-    { value: "length_lt", label: "Length less than" },
-    { value: "length_gte", label: "Length greater or equal" },
-    { value: "length_lte", label: "Length less or equal" },
-  ]},
-]
-
-const UNARY_OPERATORS = new Set(["exists", "does_not_exist", "is_empty", "is_not_empty", "is_true", "is_false"])
 
 /** Close a popout window and clear its state. Use for Save/Cancel buttons — NOT for onClose (which fires from beforeunload when the popup is already closing). */
 function closePopout(popup: Window | null, setter: (w: Window | null) => void) {
@@ -84,9 +39,6 @@ function closePopout(popup: Window | null, setter: (w: Window | null) => void) {
   setter(null)
 }
 
-function generateRuleId(): string {
-  return "r_" + Math.random().toString(36).slice(2, 8)
-}
 
 function formatTimestamp(ts: string | undefined): string {
   if (!ts) return ""
@@ -110,19 +62,6 @@ export default function NodeDetailsPanel({ slug, node, workflow, onClose }: Prop
   return <NodeConfigPanel key={node.node_id} slug={slug} node={node} workflow={workflow} onClose={onClose} />
 }
 
-/** Parse a full field path like "node_outputs.cat_1.category" into { sourceNodeId, outputField }. */
-function parseFieldPath(field: string): { sourceNodeId: string; outputField: string } {
-  if (!field || !field.startsWith("node_outputs.")) return { sourceNodeId: "", outputField: field }
-  const rest = field.slice("node_outputs.".length) // "cat_1.category"
-  const dotIdx = rest.indexOf(".")
-  if (dotIdx === -1) return { sourceNodeId: rest, outputField: "" }
-  return { sourceNodeId: rest.slice(0, dotIdx), outputField: rest.slice(dotIdx + 1) }
-}
-
-function buildFieldPath(sourceNodeId: string, outputField: string): string {
-  if (!sourceNodeId) return outputField
-  return `node_outputs.${sourceNodeId}${outputField ? "." + outputField : ""}`
-}
 
 function NodeConfigPanel({ slug, node, workflow, onClose }: Props) {
   const updateNode = useUpdateNode(slug)
@@ -206,6 +145,11 @@ function NodeConfigPanel({ slug, node, workflow, onClose }: Props) {
   const [filterSourceNode, setFilterSourceNode] = useState<string>((node.config.extra_config?.source_node as string) ?? "")
   const [filterField, setFilterField] = useState<string>((node.config.extra_config?.field as string) ?? "")
 
+  // Assertion state
+  const [assertionRules, setAssertionRules] = useState<FilterRule[]>(() => (node.config.extra_config?.rules as FilterRule[]) ?? [])
+  const [assertionJudge, setAssertionJudge] = useState<boolean>(Boolean(node.config.extra_config?.use_llm_judge))
+  const [assertionThreshold, setAssertionThreshold] = useState<number>((node.config.extra_config?.pass_threshold as number) ?? 1.0)
+
   // Merge state
   const [mergeMode, setMergeMode] = useState<string>((node.config.extra_config?.mode as string) ?? "append")
 
@@ -390,6 +334,9 @@ function NodeConfigPanel({ slug, node, workflow, onClose }: Props) {
     if (node.component_type === "filter") {
       parsedExtra = { ...parsedExtra, rules: filterRules, source_node: filterSourceNode || undefined, field: filterField || undefined }
     }
+    if (node.component_type === "assertion") {
+      parsedExtra = { ...parsedExtra, rules: assertionRules, use_llm_judge: assertionJudge, pass_threshold: assertionThreshold }
+    }
     if (node.component_type === "merge") {
       parsedExtra = { ...parsedExtra, mode: mergeMode }
     }
@@ -1375,119 +1322,18 @@ function NodeConfigPanel({ slug, node, workflow, onClose }: Props) {
       {node.component_type === "switch" && (
         <>
           <Separator />
-          <div className="space-y-3">
-            <div className="flex items-center justify-between">
-              <Label className="text-xs font-semibold">Routing Rules</Label>
-              <Button
-                variant="outline"
-                size="sm"
-                className="h-6 px-2 text-xs"
-                onClick={() => {
-                  const defaultField = upstreamNodes.length === 1 ? buildFieldPath(upstreamNodes[0], "") : ""
-                  setSwitchRules((prev) => [...prev, { id: generateRuleId(), field: defaultField, operator: "equals", value: "", label: "" }])
-                }}
-              >
-                <Plus className="h-3 w-3 mr-1" />
-                Add Rule
-              </Button>
-            </div>
-            {switchRules.length === 0 && (
-              <p className="text-xs text-muted-foreground text-center py-2">No rules defined. Add a rule to create routing branches.</p>
-            )}
-            {switchRules.map((rule, idx) => (
-              <div key={rule.id} className="border rounded-md p-2 space-y-2">
-                <div className="flex items-center justify-between">
-                  <span className="text-[10px] text-muted-foreground">Rule {idx + 1}</span>
-                  <Button
-                    variant="ghost"
-                    size="sm"
-                    className="h-5 w-5 p-0"
-                    onClick={() => setSwitchRules((prev) => prev.filter((r) => r.id !== rule.id))}
-                  >
-                    <Trash2 className="h-3 w-3" />
-                  </Button>
-                </div>
-                <div className="space-y-1">
-                  <Label className="text-[10px]">Label</Label>
-                  <Input
-                    value={rule.label}
-                    onChange={(e) => setSwitchRules((prev) => prev.map((r) => r.id === rule.id ? { ...r, label: e.target.value } : r))}
-                    className="text-xs h-7"
-                    placeholder="e.g. Good"
-                  />
-                </div>
-                <div className="space-y-1">
-                  <Label className="text-[10px]">Source Node</Label>
-                  {upstreamNodes.length > 0 ? (
-                    <Select
-                      value={parseFieldPath(rule.field).sourceNodeId || (upstreamNodes.length === 1 ? upstreamNodes[0] : "")}
-                      onValueChange={(v) => setSwitchRules((prev) => prev.map((r) => r.id === rule.id ? { ...r, field: buildFieldPath(v, parseFieldPath(r.field).outputField) } : r))}
-                    >
-                      <SelectTrigger className="text-xs h-7 font-mono"><SelectValue placeholder="Select source node" /></SelectTrigger>
-                      <SelectContent>
-                        {upstreamNodes.map((nid) => (
-                          <SelectItem key={nid} value={nid} className="text-xs font-mono">{nid}</SelectItem>
-                        ))}
-                      </SelectContent>
-                    </Select>
-                  ) : (
-                    <Input
-                      value={parseFieldPath(rule.field).sourceNodeId}
-                      onChange={(e) => setSwitchRules((prev) => prev.map((r) => r.id === rule.id ? { ...r, field: buildFieldPath(e.target.value, parseFieldPath(r.field).outputField) } : r))}
-                      className="text-xs h-7 font-mono"
-                      placeholder="node_id"
-                    />
-                  )}
-                </div>
-                <div className="space-y-1">
-                  <Label className="text-[10px]">Output Field</Label>
-                  <Input
-                    value={parseFieldPath(rule.field).outputField}
-                    onChange={(e) => setSwitchRules((prev) => prev.map((r) => r.id === rule.id ? { ...r, field: buildFieldPath(parseFieldPath(r.field).sourceNodeId, e.target.value) } : r))}
-                    className="text-xs h-7 font-mono"
-                    placeholder="e.g. category"
-                  />
-                </div>
-                <div className="space-y-1">
-                  <Label className="text-[10px]">Operator</Label>
-                  <Select
-                    value={rule.operator}
-                    onValueChange={(v) => setSwitchRules((prev) => prev.map((r) => r.id === rule.id ? { ...r, operator: v } : r))}
-                  >
-                    <SelectTrigger className="text-xs h-7"><SelectValue /></SelectTrigger>
-                    <SelectContent>
-                      {OPERATOR_OPTIONS.map((group) => (
-                        <div key={group.group}>
-                          <div className="px-2 py-1 text-[10px] font-semibold text-muted-foreground">{group.group}</div>
-                          {group.options.map((op) => (
-                            <SelectItem key={op.value} value={op.value} className="text-xs">{op.label}</SelectItem>
-                          ))}
-                        </div>
-                      ))}
-                    </SelectContent>
-                  </Select>
-                </div>
-                {!UNARY_OPERATORS.has(rule.operator) && (
-                  <div className="space-y-1">
-                    <Label className="text-[10px]">Value</Label>
-                    <Input
-                      value={rule.value}
-                      onChange={(e) => setSwitchRules((prev) => prev.map((r) => r.id === rule.id ? { ...r, value: e.target.value } : r))}
-                      className="text-xs h-7"
-                      placeholder="comparison value"
-                    />
-                  </div>
-                )}
-              </div>
-            ))}
-            <div className="flex items-center justify-between pt-1">
-              <div>
-                <Label className="text-xs">Fallback Route</Label>
-                <p className="text-[10px] text-muted-foreground">Route to "other" when no rules match</p>
-              </div>
-              <Switch checked={enableFallback} onCheckedChange={setEnableFallback} />
-            </div>
-          </div>
+          <RuleEditor<SwitchRule>
+            rules={switchRules}
+            onChange={setSwitchRules}
+            upstreamNodes={upstreamNodes}
+            showLabel
+            showSourceNode
+            showFallback
+            enableFallback={enableFallback}
+            onFallbackChange={setEnableFallback}
+            title="Routing Rules"
+            emptyMessage="No rules defined. Add a rule to create routing branches."
+          />
         </>
       )}
 
@@ -1543,75 +1389,52 @@ function NodeConfigPanel({ slug, node, workflow, onClose }: Props) {
               <Label className="text-[10px]">Source Field (optional)</Label>
               <Input value={filterField} onChange={(e) => setFilterField(e.target.value)} className="text-xs h-7 font-mono" placeholder="e.g. items" />
             </div>
+            <RuleEditor<FilterRule>
+              rules={filterRules}
+              onChange={setFilterRules}
+              upstreamNodes={upstreamNodes}
+              emptyMessage="No rules defined. All items will pass through."
+            />
+          </div>
+        </>
+      )}
+
+      {node.component_type === "assertion" && (
+        <>
+          <Separator />
+          <RuleEditor<FilterRule>
+            rules={assertionRules}
+            onChange={setAssertionRules}
+            upstreamNodes={upstreamNodes}
+            title="Assertion Rules"
+            emptyMessage="No rules defined. Add rules to validate data."
+          />
+          <Separator />
+          <div className="space-y-3">
             <div className="flex items-center justify-between">
-              <Label className="text-xs font-semibold">Rules</Label>
-              <Button
-                variant="outline"
-                size="sm"
-                className="h-6 px-2 text-xs"
-                onClick={() => setFilterRules((prev) => [...prev, { id: generateRuleId(), field: "", operator: "equals", value: "" }])}
-              >
-                <Plus className="h-3 w-3 mr-1" />
-                Add Rule
-              </Button>
+              <div>
+                <Label className="text-xs">LLM Judge</Label>
+                <p className="text-[10px] text-muted-foreground">Use an LLM to evaluate assertions</p>
+              </div>
+              <Switch checked={assertionJudge} onCheckedChange={setAssertionJudge} />
             </div>
-            {filterRules.length === 0 && (
-              <p className="text-xs text-muted-foreground text-center py-2">No rules defined. All items will pass through.</p>
-            )}
-            {filterRules.map((rule, idx) => (
-              <div key={rule.id} className="border rounded-md p-2 space-y-2">
-                <div className="flex items-center justify-between">
-                  <span className="text-[10px] text-muted-foreground">Rule {idx + 1}</span>
-                  <Button
-                    variant="ghost"
-                    size="sm"
-                    className="h-5 w-5 p-0"
-                    onClick={() => setFilterRules((prev) => prev.filter((r) => r.id !== rule.id))}
-                  >
-                    <Trash2 className="h-3 w-3" />
-                  </Button>
-                </div>
-                <div className="space-y-1">
-                  <Label className="text-[10px]">Field</Label>
-                  <Input
-                    value={rule.field}
-                    onChange={(e) => setFilterRules((prev) => prev.map((r) => r.id === rule.id ? { ...r, field: e.target.value } : r))}
-                    className="text-xs h-7 font-mono"
-                    placeholder="e.g. name, status"
+            {assertionJudge && (
+              <div className="space-y-1">
+                <Label className="text-[10px]">Pass Threshold</Label>
+                <div className="flex items-center gap-2">
+                  <input
+                    type="range"
+                    min="0"
+                    max="1"
+                    step="0.05"
+                    value={assertionThreshold}
+                    onChange={(e) => setAssertionThreshold(Number(e.target.value))}
+                    className="flex-1 h-1.5 accent-indigo-500"
                   />
+                  <span className="text-xs font-mono w-10 text-right">{assertionThreshold.toFixed(2)}</span>
                 </div>
-                <div className="space-y-1">
-                  <Label className="text-[10px]">Operator</Label>
-                  <Select
-                    value={rule.operator}
-                    onValueChange={(v) => setFilterRules((prev) => prev.map((r) => r.id === rule.id ? { ...r, operator: v } : r))}
-                  >
-                    <SelectTrigger className="text-xs h-7"><SelectValue /></SelectTrigger>
-                    <SelectContent>
-                      {OPERATOR_OPTIONS.map((group) => (
-                        <div key={group.group}>
-                          <div className="px-2 py-1 text-[10px] font-semibold text-muted-foreground">{group.group}</div>
-                          {group.options.map((op) => (
-                            <SelectItem key={op.value} value={op.value} className="text-xs">{op.label}</SelectItem>
-                          ))}
-                        </div>
-                      ))}
-                    </SelectContent>
-                  </Select>
-                </div>
-                {!UNARY_OPERATORS.has(rule.operator) && (
-                  <div className="space-y-1">
-                    <Label className="text-[10px]">Value</Label>
-                    <Input
-                      value={rule.value}
-                      onChange={(e) => setFilterRules((prev) => prev.map((r) => r.id === rule.id ? { ...r, value: e.target.value } : r))}
-                      className="text-xs h-7"
-                      placeholder="comparison value"
-                    />
-                  </div>
-                )}
               </div>
-            ))}
+            )}
           </div>
         </>
       )}
diff --git a/platform/frontend/src/features/workflows/components/NodePalette.tsx b/platform/frontend/src/features/workflows/components/NodePalette.tsx
index 45bfcd7..0ef7211 100644
--- a/platform/frontend/src/features/workflows/components/NodePalette.tsx
+++ b/platform/frontend/src/features/workflows/components/NodePalette.tsx
@@ -7,7 +7,7 @@ import {
   Cpu, Bot, Brain, GraduationCap,
   GitFork, Route, FileOutput, Split,
   Terminal,
-  Repeat, Pause, Merge, Filter,
+  Repeat, Pause, Merge, Filter, ClipboardCheck,
   Code, UserCheck, ShieldAlert, FileText, CheckSquare, FileCheck,
   Database, DatabaseZap, UserSearch, UserPlus, Plug, Fingerprint, KeyRound,
   Rocket, PencilRuler, CalendarClock, HeartPulse,
@@ -54,6 +54,7 @@ const ICONS: Record<ComponentType, LucideIcon> = {
   reply_chat: MessageSquare,
   validate_gherkin: CheckSquare,
   validate_topology: FileCheck,
+  assertion: ClipboardCheck,
 }
 
 const NODE_CATEGORIES: { label: string; types: ComponentType[] }[] = [
@@ -63,7 +64,7 @@ const NODE_CATEGORIES: { label: string; types: ComponentType[] }[] = [
   { label: "Memory", types: ["memory_read", "memory_write", "identify_user"] },
   { label: "Agent", types: ["whoami", "create_agent_user", "get_totp_code", "platform_api", "scheduler_tools", "system_health", "spawn_and_await", "workflow_create"] },
   { label: "Tools", types: ["run_command", "workflow_discover", "validate_gherkin", "validate_topology"] },
-  { label: "Logic", types: ["switch", "loop", "filter", "merge", "wait"] },
+  { label: "Logic", types: ["switch", "loop", "filter", "merge", "wait", "assertion"] },
   { label: "Output", types: ["reply_chat"] },
   { label: "Other", types: ["workflow", "code", "human_confirmation", "error_handler", "output_parser"] },
 ]
diff --git a/platform/frontend/src/features/workflows/components/RuleEditor.tsx b/platform/frontend/src/features/workflows/components/RuleEditor.tsx
new file mode 100644
index 0000000..a44ffd3
--- /dev/null
+++ b/platform/frontend/src/features/workflows/components/RuleEditor.tsx
@@ -0,0 +1,248 @@
+import { Button } from "@/components/ui/button"
+import { Input } from "@/components/ui/input"
+import { Label } from "@/components/ui/label"
+import { Switch } from "@/components/ui/switch"
+import { Select, SelectContent, SelectItem, SelectTrigger, SelectValue } from "@/components/ui/select"
+import { Trash2, Plus } from "lucide-react"
+import type { Rule, SwitchRule } from "@/types/models"
+
+const OPERATOR_OPTIONS = [
+  { group: "Universal", options: [
+    { value: "exists", label: "Exists" },
+    { value: "does_not_exist", label: "Does not exist" },
+    { value: "is_empty", label: "Is empty" },
+    { value: "is_not_empty", label: "Is not empty" },
+  ]},
+  { group: "String", options: [
+    { value: "equals", label: "Equals" },
+    { value: "not_equals", label: "Not equals" },
+    { value: "contains", label: "Contains" },
+    { value: "not_contains", label: "Not contains" },
+    { value: "starts_with", label: "Starts with" },
+    { value: "not_starts_with", label: "Not starts with" },
+    { value: "ends_with", label: "Ends with" },
+    { value: "not_ends_with", label: "Not ends with" },
+    { value: "matches_regex", label: "Matches regex" },
+    { value: "not_matches_regex", label: "Not matches regex" },
+  ]},
+  { group: "Number", options: [
+    { value: "gt", label: "Greater than" },
+    { value: "lt", label: "Less than" },
+    { value: "gte", label: "Greater or equal" },
+    { value: "lte", label: "Less or equal" },
+  ]},
+  { group: "Datetime", options: [
+    { value: "after", label: "After" },
+    { value: "before", label: "Before" },
+    { value: "after_or_equal", label: "After or equal" },
+    { value: "before_or_equal", label: "Before or equal" },
+  ]},
+  { group: "Boolean", options: [
+    { value: "is_true", label: "Is true" },
+    { value: "is_false", label: "Is false" },
+  ]},
+  { group: "Array", options: [
+    { value: "length_eq", label: "Length equals" },
+    { value: "length_neq", label: "Length not equals" },
+    { value: "length_gt", label: "Length greater than" },
+    { value: "length_lt", label: "Length less than" },
+    { value: "length_gte", label: "Length greater or equal" },
+    { value: "length_lte", label: "Length less or equal" },
+  ]},
+]
+
+const UNARY_OPERATORS = new Set(["exists", "does_not_exist", "is_empty", "is_not_empty", "is_true", "is_false"])
+
+export function generateRuleId(): string {
+  return "r_" + Math.random().toString(36).slice(2, 11) + Date.now().toString(36)
+}
+
+/** Parse a full field path like "node_outputs.cat_1.category" into { sourceNodeId, outputField }. */
+export function parseFieldPath(field: string): { sourceNodeId: string; outputField: string } {
+  if (!field || !field.startsWith("node_outputs.")) return { sourceNodeId: "", outputField: field }
+  const rest = field.slice("node_outputs.".length)
+  const dotIdx = rest.indexOf(".")
+  if (dotIdx === -1) return { sourceNodeId: rest, outputField: "" }
+  return { sourceNodeId: rest.slice(0, dotIdx), outputField: rest.slice(dotIdx + 1) }
+}
+
+export function buildFieldPath(sourceNodeId: string, outputField: string): string {
+  if (!sourceNodeId) return outputField
+  return `node_outputs.${sourceNodeId}${outputField ? "." + outputField : ""}`
+}
+
+interface RuleEditorProps<T extends Rule> {
+  rules: T[]
+  onChange: (rules: T[]) => void
+  upstreamNodes: string[]
+  /** Show per-rule label field (switch mode) */
+  showLabel?: boolean
+  /** Show source node dropdown per rule (switch mode — source is part of field path) */
+  showSourceNode?: boolean
+  /** Show fallback toggle */
+  showFallback?: boolean
+  enableFallback?: boolean
+  onFallbackChange?: (v: boolean) => void
+  /** Title for the rules section */
+  title?: string
+  /** Empty state message */
+  emptyMessage?: string
+}
+
+export default function RuleEditor<T extends Rule>({
+  rules,
+  onChange,
+  upstreamNodes,
+  showLabel = false,
+  showSourceNode = false,
+  showFallback = false,
+  enableFallback = false,
+  onFallbackChange,
+  title = "Rules",
+  emptyMessage = "No rules defined.",
+}: RuleEditorProps<T>) {
+  const addRule = () => {
+    const defaultField = showSourceNode && upstreamNodes.length === 1 ? buildFieldPath(upstreamNodes[0], "") : ""
+    const newRule = { id: generateRuleId(), field: defaultField, operator: "equals", value: "", ...(showLabel ? { label: "" } : {}) } as T
+    onChange([...rules, newRule])
+  }
+
+  const removeRule = (id: string) => {
+    onChange(rules.filter((r) => r.id !== id))
+  }
+
+  const updateRule = (id: string, patch: Partial<T>) => {
+    onChange(rules.map((r) => r.id === id ? { ...r, ...patch } : r))
+  }
+
+  return (
+    <div className="space-y-3">
+      <div className="flex items-center justify-between">
+        <Label className="text-xs font-semibold">{title}</Label>
+        <Button
+          variant="outline"
+          size="sm"
+          className="h-6 px-2 text-xs"
+          onClick={addRule}
+        >
+          <Plus className="h-3 w-3 mr-1" />
+          Add Rule
+        </Button>
+      </div>
+      {rules.length === 0 && (
+        <p className="text-xs text-muted-foreground text-center py-2">{emptyMessage}</p>
+      )}
+      {rules.map((rule, idx) => (
+        <div key={rule.id} className="border rounded-md p-2 space-y-2">
+          <div className="flex items-center justify-between">
+            <span className="text-[10px] text-muted-foreground">Rule {idx + 1}</span>
+            <Button
+              variant="ghost"
+              size="sm"
+              className="h-5 w-5 p-0"
+              onClick={() => removeRule(rule.id)}
+            >
+              <Trash2 className="h-3 w-3" />
+            </Button>
+          </div>
+          {showLabel && (
+            <div className="space-y-1">
+              <Label className="text-[10px]">Label</Label>
+              <Input
+                value={(rule as unknown as SwitchRule).label ?? ""}
+                onChange={(e) => updateRule(rule.id, { label: e.target.value } as unknown as Partial<T>)}
+                className="text-xs h-7"
+                placeholder="e.g. Good"
+              />
+            </div>
+          )}
+          {showSourceNode ? (
+            <>
+              <div className="space-y-1">
+                <Label className="text-[10px]">Source Node</Label>
+                {upstreamNodes.length > 0 ? (
+                  <Select
+                    value={parseFieldPath(rule.field).sourceNodeId || (upstreamNodes.length === 1 ? upstreamNodes[0] : "")}
+                    onValueChange={(v) => updateRule(rule.id, { field: buildFieldPath(v, parseFieldPath(rule.field).outputField) } as Partial<T>)}
+                  >
+                    <SelectTrigger className="text-xs h-7 font-mono"><SelectValue placeholder="Select source node" /></SelectTrigger>
+                    <SelectContent>
+                      {upstreamNodes.map((nid) => (
+                        <SelectItem key={nid} value={nid} className="text-xs font-mono">{nid}</SelectItem>
+                      ))}
+                    </SelectContent>
+                  </Select>
+                ) : (
+                  <Input
+                    value={parseFieldPath(rule.field).sourceNodeId}
+                    onChange={(e) => updateRule(rule.id, { field: buildFieldPath(e.target.value, parseFieldPath(rule.field).outputField) } as Partial<T>)}
+                    className="text-xs h-7 font-mono"
+                    placeholder="node_id"
+                  />
+                )}
+              </div>
+              <div className="space-y-1">
+                <Label className="text-[10px]">Output Field</Label>
+                <Input
+                  value={parseFieldPath(rule.field).outputField}
+                  onChange={(e) => updateRule(rule.id, { field: buildFieldPath(parseFieldPath(rule.field).sourceNodeId, e.target.value) } as Partial<T>)}
+                  className="text-xs h-7 font-mono"
+                  placeholder="e.g. category"
+                />
+              </div>
+            </>
+          ) : (
+            <div className="space-y-1">
+              <Label className="text-[10px]">Field</Label>
+              <Input
+                value={rule.field}
+                onChange={(e) => updateRule(rule.id, { field: e.target.value } as Partial<T>)}
+                className="text-xs h-7 font-mono"
+                placeholder="e.g. name, status"
+              />
+            </div>
+          )}
+          <div className="space-y-1">
+            <Label className="text-[10px]">Operator</Label>
+            <Select
+              value={rule.operator}
+              onValueChange={(v) => updateRule(rule.id, { operator: v } as Partial<T>)}
+            >
+              <SelectTrigger className="text-xs h-7"><SelectValue /></SelectTrigger>
+              <SelectContent>
+                {OPERATOR_OPTIONS.map((group) => (
+                  <div key={group.group}>
+                    <div className="px-2 py-1 text-[10px] font-semibold text-muted-foreground">{group.group}</div>
+                    {group.options.map((op) => (
+                      <SelectItem key={op.value} value={op.value} className="text-xs">{op.label}</SelectItem>
+                    ))}
+                  </div>
+                ))}
+              </SelectContent>
+            </Select>
+          </div>
+          {!UNARY_OPERATORS.has(rule.operator) && (
+            <div className="space-y-1">
+              <Label className="text-[10px]">Value</Label>
+              <Input
+                value={rule.value}
+                onChange={(e) => updateRule(rule.id, { value: e.target.value } as Partial<T>)}
+                className="text-xs h-7"
+                placeholder="comparison value"
+              />
+            </div>
+          )}
+        </div>
+      ))}
+      {showFallback && (
+        <div className="flex items-center justify-between pt-1">
+          <div>
+            <Label className="text-xs">Fallback Route</Label>
+            <p className="text-[10px] text-muted-foreground">Route to "other" when no rules match</p>
+          </div>
+          <Switch checked={enableFallback} onCheckedChange={onFallbackChange} />
+        </div>
+      )}
+    </div>
+  )
+}
diff --git a/platform/frontend/src/features/workflows/components/WorkflowCanvas.tsx b/platform/frontend/src/features/workflows/components/WorkflowCanvas.tsx
index 4d95091..f3c3655 100644
--- a/platform/frontend/src/features/workflows/components/WorkflowCanvas.tsx
+++ b/platform/frontend/src/features/workflows/components/WorkflowCanvas.tsx
@@ -4,7 +4,7 @@ import { FontAwesomeIcon } from "@fortawesome/react-fontawesome"
 import type { IconDefinition } from "@fortawesome/fontawesome-svg-core"
 import {
   faMicrochip, faRobot, faTags, faCodeBranch, faWrench, faMagnifyingGlassChart,
-  faSitemap, faCode, faTriangleExclamation, faUserCheck,
+  faSitemap, faCode, faTriangleExclamation, faUserCheck, faClipboardCheck,
   faFileExport, faRepeat, faClock, faCodeMerge, faFilter,
   faCalendarDays, faHandPointer, faHourglass, faHeartPulse,
   faPlay, faBug, faComments, faCircleNotch, faCircleCheck, faCircleXmark, faMinus,
@@ -81,6 +81,7 @@ const COMPONENT_COLORS: Record<string, string> = {
   filter: "#6366f1",
   merge: "#6366f1",
   wait: "#6366f1",
+  assertion: "#6366f1",
   workflow: "#6366f1",
   code: "#64748b",
   memory_read: "#f59e0b",
@@ -108,7 +109,7 @@ const COMPONENT_ICONS: Record<string, IconDefinition> = {
   code: faCode, error_handler: faTriangleExclamation,
   memory_read: faDatabase, memory_write: faFloppyDisk, identify_user: faIdCard,
   human_confirmation: faUserCheck, output_parser: faFileExport,
-  loop: faRepeat, wait: faClock, merge: faCodeMerge, filter: faFilter,
+  loop: faRepeat, wait: faClock, merge: faCodeMerge, filter: faFilter, assertion: faClipboardCheck,
   trigger_telegram: faTelegram, trigger_schedule: faCalendarDays,
   trigger_manual: faHandPointer, trigger_workflow: faPlay, trigger_error: faBug,
   trigger_chat: faComments,
@@ -129,7 +130,7 @@ function WorkflowNodeComponent({ data, selected }: { data: { label: string; comp
   const isWaiting = data.executionStatus === "waiting"
   const isTrigger = data.componentType.startsWith("trigger_")
   const isLoop = data.componentType === "loop"
-  const isFixedWidth = ["router", "categorizer", "agent", "deep_agent", "extractor", "switch", "loop"].includes(data.componentType)
+  const isFixedWidth = ["router", "categorizer", "agent", "deep_agent", "extractor", "switch", "loop", "assertion"].includes(data.componentType)
   const isTool = ["run_command", "memory_read", "memory_write", "create_agent_user", "platform_api", "whoami", "scheduler_tools", "system_health", "spawn_and_await", "workflow_create", "workflow_discover", "validate_gherkin", "validate_topology"].includes(data.componentType)
   const isSubComponent = ["ai_model", "run_command", "output_parser", "memory_read", "memory_write", "create_agent_user", "platform_api", "whoami", "scheduler_tools", "system_health", "spawn_and_await", "workflow_create", "workflow_discover", "skill", "validate_gherkin", "validate_topology"].includes(data.componentType)
   const isAiModel = data.componentType === "ai_model"
@@ -146,6 +147,7 @@ function WorkflowNodeComponent({ data, selected }: { data: { label: string; comp
   const isSuccess = data.executionStatus === "success"
   const isFailed = data.executionStatus === "failed"
   const isSwitch = data.componentType === "switch"
+  const isAssertion = data.componentType === "assertion"
   const switchHandles = isSwitch ? (data.rules ?? []) : []
   const showFallbackHandle = isSwitch && data.enableFallback
   return (
@@ -210,8 +212,8 @@ function WorkflowNodeComponent({ data, selected }: { data: { label: string; comp
           </PopoverContent>
         </Popover>
       )}
-      {isFixedWidth && !isSwitch && !isLoop && <hr className="border-muted-foreground/30 my-1" />}
-      {isFixedWidth && !isSwitch && !isLoop && (
+      {isFixedWidth && !isSwitch && !isLoop && !isAssertion && <hr className="border-muted-foreground/30 my-1" />}
+      {isFixedWidth && !isSwitch && !isLoop && !isAssertion && (
         <div className="flex mt-1">
           {hasModel && (
             <div className="relative p-1.5 bg-background rounded-[10px]" style={{ color: "#3b82f6", borderColor: "#3b82f6", borderWidth: 1, borderStyle: "solid" }} title="model">
@@ -283,7 +285,22 @@ function WorkflowNodeComponent({ data, selected }: { data: { label: string; comp
           />
         </>
       )}
-      {!isSubComponent && !(isSwitch && (switchHandles.length > 0 || showFallbackHandle)) && !isLoop && <Handle type="source" position={Position.Right} className="!bg-muted-foreground !w-2 !h-2" />}
+      {isAssertion && (
+        <>
+          <hr className="border-muted-foreground/30 my-1" />
+          <div className="flex flex-col gap-1">
+            <div className="relative flex items-center justify-end pr-4">
+              <span className="text-[10px] text-emerald-500 font-medium">Pass</span>
+              <Handle type="source" position={Position.Right} id="pass" className="!bg-emerald-500 !w-2 !h-2" />
+            </div>
+            <div className="relative flex items-center justify-end pr-4">
+              <span className="text-[10px] text-red-500 font-medium">Fail</span>
+              <Handle type="source" position={Position.Right} id="fail" className="!bg-red-500 !w-2 !h-2" />
+            </div>
+          </div>
+        </>
+      )}
+      {!isSubComponent && !(isSwitch && (switchHandles.length > 0 || showFallbackHandle)) && !isLoop && !isAssertion && <Handle type="source" position={Position.Right} className="!bg-muted-foreground !w-2 !h-2" />}
     </div>
   )
 }
@@ -450,6 +467,8 @@ export default function WorkflowCanvas({ slug, workflow, selectedNodeId, onSelec
         const rules = (srcNode.config?.extra_config?.rules as SwitchRule[]) ?? []
         const rule = rules.find((r) => r.id === e.condition_value)
         condLabel = rule?.label || e.condition_value
+      } else if (srcNode?.component_type === "assertion") {
+        condLabel = e.condition_value // "pass" or "fail"
       } else {
         condLabel = e.condition_value
       }
@@ -514,6 +533,16 @@ export default function WorkflowCanvas({ slug, workflow, selectedNodeId, onSelec
         })
         return
       }
+      if (sourceNode?.component_type === "assertion" && !edge_label && (params.sourceHandle === "pass" || params.sourceHandle === "fail")) {
+        createEdge.mutate({
+          source_node_id: params.source,
+          target_node_id: params.target,
+          edge_type: "conditional",
+          edge_label,
+          condition_value: params.sourceHandle,
+        })
+        return
+      }
       if (sourceNode?.component_type === "switch" && !edge_label) {
         // If dragged from a specific rule handle, auto-create conditional edge
         const ruleId = params.sourceHandle
diff --git a/platform/frontend/src/types/models.ts b/platform/frontend/src/types/models.ts
index a9aadb3..644d15c 100644
--- a/platform/frontend/src/types/models.ts
+++ b/platform/frontend/src/types/models.ts
@@ -39,6 +39,7 @@ export type ComponentType =
   | "skill"
   | "validate_gherkin"
   | "validate_topology"
+  | "assertion"
 export type EdgeType = "direct" | "conditional"
 // "memory" was removed — migration 0d301d48b86a converts all memory edges to tool edges.
 export type EdgeLabel = "" | "llm" | "tool" | "output_parser" | "loop_body" | "loop_return" | "skill"
@@ -112,11 +113,10 @@ export interface WorkspaceUpdate { allow_network?: boolean; env_vars?: Workspace
 // Paginated response
 export interface PaginatedResponse<T> { items: T[]; total: number }
 
-// Switch rules
-export interface SwitchRule { id: string; field: string; operator: string; value: string; label: string }
-
-// Filter rules
-export interface FilterRule { id: string; field: string; operator: string; value: string }
+// Rules (shared base for switch, filter, assertion)
+export interface Rule { id: string; field: string; operator: string; value: string }
+export interface SwitchRule extends Rule { label: string }
+export type FilterRule = Rule
 
 // Checkpoints
 export interface Checkpoint { thread_id: string; checkpoint_ns: string; checkpoint_id: string; parent_checkpoint_id: string | null; step: number | null; source: string | null; blob_size: number }
diff --git a/platform/models/node.py b/platform/models/node.py
index d7116a9..1be3795 100644
--- a/platform/models/node.py
+++ b/platform/models/node.py
@@ -125,6 +125,10 @@ class _SwitchConfig(BaseComponentConfig):
     __mapper_args__ = {"polymorphic_identity": "switch"}
 
 
+class _AssertionConfig(BaseComponentConfig):
+    __mapper_args__ = {"polymorphic_identity": "assertion"}
+
+
 class CodeComponentConfig(BaseComponentConfig):
     """Config for code-type components."""
     __mapper_args__ = {"polymorphic_identity": "code"}
@@ -263,6 +267,7 @@ class _ReplyChatConfig(BaseComponentConfig):
     "router": AIComponentConfig,
     "extractor": AIComponentConfig,
     "switch": OtherComponentConfig,
+    "assertion": _AssertionConfig,
     "code": CodeComponentConfig,
     "loop": CodeComponentConfig,
     "filter": CodeComponentConfig,
diff --git a/platform/schemas/node.py b/platform/schemas/node.py
index d53ba7c..5b76b91 100644
--- a/platform/schemas/node.py
+++ b/platform/schemas/node.py
@@ -13,6 +13,7 @@
     "agent",
     "deep_agent",
     "switch",
+    "assertion",
     "run_command",
     "get_totp_code",
     "platform_api",
diff --git a/platform/schemas/node_type_defs.py b/platform/schemas/node_type_defs.py
index 9bf06e8..627160a 100644
--- a/platform/schemas/node_type_defs.py
+++ b/platform/schemas/node_type_defs.py
@@ -458,6 +458,18 @@
     outputs=[PortDefinition(name="route", data_type=DataType.STRING)],
 ))
 
+register_node_type(NodeTypeSpec(
+    component_type="assertion",
+    display_name="Assertion",
+    description="Evaluates all rules against state and routes pass/fail",
+    category="logic",
+    inputs=[PortDefinition(name="input", data_type=DataType.ANY, required=True)],
+    outputs=[
+        PortDefinition(name="output", data_type=DataType.OBJECT, description="Results with passed flag and per-rule details"),
+        PortDefinition(name="route", data_type=DataType.STRING, description="'pass' or 'fail'"),
+    ],
+))
+
 register_node_type(NodeTypeSpec(
     component_type="code",
     display_name="Code",
diff --git a/platform/services/dsl_compiler.py b/platform/services/dsl_compiler.py
index db34877..b79e4ee 100644
--- a/platform/services/dsl_compiler.py
+++ b/platform/services/dsl_compiler.py
@@ -44,6 +44,7 @@
     "agent": "agent",
     "deep_agent": "deep_agent",
     "switch": "switch",
+    "assertion": "assertion",
     "loop": "loop",
     "workflow": "workflow",
     "human": "human_confirmation",
@@ -573,6 +574,20 @@ def _build_step_config(
             "channel": step.get("channel", ""),
         }
 
+    elif step_type == "assertion":
+        rules = []
+        for rule in step.get("rules", []):
+            rules.append({
+                "id": rule.get("id", ""),
+                "field": rule.get("field", ""),
+                "operator": rule.get("operator", "equals"),
+                "value": rule.get("value", ""),
+            })
+        config["extra_config"] = {
+            "rules": rules,
+            "pass_threshold": step.get("pass_threshold", 1.0),
+        }
+
     elif step_type == "switch":
         rules = []
         for rule in step.get("rules", []):
diff --git a/platform/tests/dsl_fixtures/workflow_generator/fixture.json b/platform/tests/dsl_fixtures/workflow_generator/fixture.json
index 979e7dc..acf3d26 100644
--- a/platform/tests/dsl_fixtures/workflow_generator/fixture.json
+++ b/platform/tests/dsl_fixtures/workflow_generator/fixture.json
@@ -11,8 +11,8 @@
     "max_execution_seconds": 900,
     "input_schema": null,
     "output_schema": null,
-    "node_count": 20,
-    "edge_count": 19,
+    "node_count": 22,
+    "edge_count": 21,
     "created_at": "2026-03-20T09:01:27",
     "updated_at": "2026-03-20T09:01:27",
     "nodes": [
@@ -25,7 +25,7 @@
             "interrupt_before": false,
             "interrupt_after": false,
             "position_x": 1486,
-            "position_y": -5,
+            "position_y": -17,
             "config": {
                 "system_prompt": "You are the Builder. The topology and behavior spec have been verified. Now create the actual workflow in Pipelit using the platform API.\n\nYou have access to the platform_api and workflow_create tools. Use them to:\n1. Create the workflow (POST /api/v1/workflows/)\n2. Create each node from the topology (POST /api/v1/workflows/{slug}/nodes/)\n3. Create edges between nodes (POST /api/v1/workflows/{slug}/edges/)\n4. For agent/deep_agent nodes: create ai_model sub-components and connect via llm edges\n5. Validate the workflow (POST /api/v1/workflows/{slug}/validate/)\n\nThe LLM credential ID is 1. The model is claude-sonnet-4-20250514.\n\nTopology: {{ topology_agent.output }}\nBehavior spec: {{ gherkin_agent.output }}\n\nCreate the workflow now. Report success or any errors encountered.",
                 "extra_config": {},
@@ -48,7 +48,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T09:01:27",
+            "updated_at": "2026-03-20T21:52:12",
             "schedule_job": null
         },
         {
@@ -59,8 +59,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 1562,
-            "position_y": 150,
+            "position_x": 1532,
+            "position_y": 151,
             "config": {
                 "system_prompt": "",
                 "extra_config": {},
@@ -83,7 +83,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T09:01:27",
+            "updated_at": "2026-03-20T21:52:13",
             "schedule_job": null
         },
         {
@@ -94,8 +94,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 1089,
-            "position_y": -18,
+            "position_x": 1094,
+            "position_y": -61,
             "config": {
                 "system_prompt": "",
                 "extra_config": {
@@ -134,7 +134,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T09:01:27",
+            "updated_at": "2026-03-20T22:01:04",
             "schedule_job": null
         },
         {
@@ -145,8 +145,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": -104,
-            "position_y": -85,
+            "position_x": -98,
+            "position_y": -113,
             "config": {
                 "system_prompt": "",
                 "extra_config": {
@@ -173,7 +173,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T09:01:27",
+            "updated_at": "2026-03-21T06:08:32",
             "schedule_job": null
         },
         {
@@ -187,7 +187,7 @@
             "position_x": 170,
             "position_y": -264,
             "config": {
-                "system_prompt": "[agent]\nname = \"Gherkin Agent\"\nrole = \"\"\"\nYou are the Gherkin Agent. You receive a requirements document from the Scribe\nand produce a Gherkin behavior specification for the described workflow.\nYour output is the execution contract for the Verifier Agent — a live LLM-driven\nworkflow that will send each scenario as a real chat request to the workflow trigger\nand match actual responses against your Then clauses. Write accordingly.\n\"\"\"\n\n[context]\n# The trigger IS the entry point — do NOT create scenarios for \"receiving\" a message\ntrigger_is_entry_point = true\n\n# reply_chat is a terminal node that sends a message back to the chat caller\nreply_chat_is_terminal = true\n\n# Scenarios test end-to-end observable behaviour via the trigger endpoint\n# Node IDs are used for traceability only — assertions are always on reply_chat output\ntest_observable_behaviour_only = true\n\n[rules]\n# Every processing step from the requirements MUST appear in at least one scenario\nfull_step_coverage = true\n\n# Use the exact node IDs from the requirements document\nuse_exact_node_ids = true\n\n# Use standard Given/When/Then format\nformat = \"Given/When/Then\"\n\n# Every scenario must be independently executable\n# No scenario may depend on state or output from a previous scenario\nscenarios_are_stateless = true\n\n# Every Then clause must assert a specific, verifiable value or condition\n# The Verifier matches Then clauses directly against raw response strings — no inference\nverifiable_assertions_only = true\n\n[rules.given_when]\n# Given defines the exact input precondition — use concrete literal values\n# When defines the action — name the exact node_id performing it\n# Given + When must compose into an unambiguous natural language chat message\n# The Verifier derives the actual message to send from these two clauses verbatim\nmust_compose_to_sendable_message = true\n\n[rules.then]\n# Then clauses must be matchable against a raw response string\n# Specify exact strings, HTTP status codes, or unambiguous structural conditions\nmust_be_raw_response_matchable = true\n\n[rules.error_scenario_requirements]\n# Error scenarios must specify the exact HTTP status code returned\nrequire_http_status_code = true\n\n# Error scenarios must specify the exact error message or response body shape\nrequire_error_message = true\n\n# Error scenarios must state whether the node retries or fails immediately\nrequire_retry_behaviour = true\n\n[rules.input_boundary_requirements]\n# Every workflow that accepts string input MUST include these boundary scenarios\n[[rules.input_boundary_requirements.cases]]\ncase   = \"empty_string\"\ninput  = '\"\"'\nreason = \"empty input must have a defined rejection or fallback behaviour\"\n\n[[rules.input_boundary_requirements.cases]]\ncase   = \"whitespace_only\"\ninput  = '\"   \"'\nreason = \"whitespace-only must not be treated as valid content\"\n\n[[rules.input_boundary_requirements.cases]]\ncase   = \"null_or_missing\"\ninput  = \"null / field absent\"\nreason = \"missing input must return HTTP 400 with a specific error message\"\n\n[[rules.input_boundary_requirements.cases]]\ncase   = \"oversized_payload\"\ninput  = \"string exceeding max token or byte limit\"\nreason = \"must return HTTP 413 or equivalent with a size exceeded message\"\n\n[[rules.input_boundary_requirements.cases]]\ncase   = \"malformed_encoding\"\ninput  = \"non-UTF-8 or broken JSON\"\nreason = \"must return HTTP 400 with a parse error message\"\n\n[[rules.then_antipatterns]]\npattern = \"should generate an? (appropriate|correct|valid|concise) .*\"\nreason  = \"adjective without qualification is untestable — specify exact value or format\"\n\n[[rules.then_antipatterns]]\npattern = \"should handle .* appropriately\"\nreason  = \"unverifiable — specify the exact output or behaviour\"\n\n[[rules.then_antipatterns]]\npattern = \"should send an? (appropriate|correct|valid) response\"\nreason  = \"must specify the exact response value or structure\"\n\n[[rules.then_antipatterns]]\npattern = \"should fail gracefully\"\nreason  = \"specify the exact HTTP code, error message, and response shape\"\n\n[[rules.then_antipatterns]]\npattern = \"should return an error\"\nreason  = \"specify HTTP 4xx/5xx code, error field value, and whether retry is expected\"\n\n[[rules.then_antipatterns]]\npattern = \"should contain relevant information\"\nreason  = \"Verifier cannot match 'relevant' — specify exact field or substring\"\n\n[[rules.then_antipatterns]]\npattern = \"should respond within .* seconds\"\nreason  = \"Verifier is not a latency tester — remove timing assertions\"\n\n[[rules.then_antipatterns]]\npattern = \"should process the (input|message)\"\nreason  = \"processing is not observable — assert on the output only\"\n\n[input]\n# Receives the full requirements document from the Scribe\nsource   = \"scribe.output\"\ntemplate = \"{{ scribe.output }}\"\n\n[output]\n# Emit ONLY a single fenced code block — no preamble, no explanation\nformat = \"single fenced code block, label: behavior.feature\"\n\n[output.structure]\n# Feature name derived from workflow PURPOSE in the requirements doc\nfeature = \"<workflow name>\"\n\n# Required scenario types per processing step:\n# 1. happy path — concrete input, exact expected output\n# 2. error case — specific failure condition, HTTP code, exact error message\n# 3. input boundary — one scenario per boundary case for every string-accepting node\n[[output.structure.scenarios]]\ntype     = \"happy_path\"\ntemplate = \"\"\"\nScenario: <happy path name>\n  Given a chat message \"<exact literal input string>\"\n  When <node_id> processes the message\n  Then <node_id> should return \"<exact expected output string>\"\n  And reply_chat should send \"<exact message>\" back to the chat caller\n\"\"\"\n\n[[output.structure.scenarios]]\ntype     = \"error_case\"\ntemplate = \"\"\"\nScenario: <error case name>\n  Given a chat message \"<exact literal input string>\"\n  When <node_id> encounters <specific failure condition>\n  Then <node_id> should return HTTP <status_code> with error \"<exact error message>\"\n  And reply_chat should send \"<exact fallback message>\" back to the chat caller\n\"\"\"\n\n[[output.structure.scenarios]]\ntype     = \"input_boundary\"\ntemplate = \"\"\"\nScenario: <boundary case name>\n  Given a <empty|whitespace|null|oversized|malformed> chat message <exact input value>\n  When <node_id> attempts to process the input\n  Then <node_id> should return HTTP <status_code> with error \"<exact error message>\"\n  And reply_chat should send \"<exact response>\" back to the chat caller\n\"\"\"\n\n[validation]\n# Before returning your output, you MUST call the validate_gherkin tool\n# to check your Gherkin spec for syntax errors and lint warnings.\n# Fix any issues the tool reports before returning your final output.\nrequire_validation = true\ntool_name = \"validate_gherkin\"\n\n[validation.workflow]\n# 1. Generate your Gherkin spec\n# 2. Call validate_gherkin with the raw spec text (no fences)\n# 3. If valid=false or lint_errors exist, fix the issues and re-validate\n# 4. Only return the spec once it passes validation\nmust_pass_before_output = true\n",
+                "system_prompt": "[agent]\nname = \"Gherkin Agent\"\nrole = \"\"\"\nYou are the Gherkin Agent. You receive a requirements document from the Scribe\nand produce a Gherkin behavior specification for the described workflow.\nYour output is the execution contract for the Verifier Agent \u2014 a live LLM-driven\nworkflow that will send each scenario as a real chat request to the workflow trigger\nand match actual responses against your Then clauses. Write accordingly.\n\"\"\"\n\n[context]\n# The trigger IS the entry point \u2014 do NOT create scenarios for \"receiving\" a message\ntrigger_is_entry_point = true\n\n# reply_chat is a terminal node that sends a message back to the chat caller\nreply_chat_is_terminal = true\n\n# Scenarios test end-to-end observable behaviour via the trigger endpoint\n# Node IDs are used for traceability only \u2014 assertions are always on reply_chat output\ntest_observable_behaviour_only = true\n\n[rules]\n# Every processing step from the requirements MUST appear in at least one scenario\nfull_step_coverage = true\n\n# Use the exact node IDs from the requirements document\nuse_exact_node_ids = true\n\n# Use standard Given/When/Then format\nformat = \"Given/When/Then\"\n\n# Every scenario must be independently executable\n# No scenario may depend on state or output from a previous scenario\nscenarios_are_stateless = true\n\n# Every Then clause must assert a specific, verifiable value or condition\n# The Verifier matches Then clauses directly against raw response strings \u2014 no inference\nverifiable_assertions_only = true\n\n[rules.given_when]\n# Given defines the exact input precondition \u2014 use concrete literal values\n# When defines the action \u2014 name the exact node_id performing it\n# Given + When must compose into an unambiguous natural language chat message\n# The Verifier derives the actual message to send from these two clauses verbatim\nmust_compose_to_sendable_message = true\n\n[rules.then]\n# Then clauses must be matchable against a raw response string\n# Specify exact strings, HTTP status codes, or unambiguous structural conditions\nmust_be_raw_response_matchable = true\n\n[rules.error_scenario_requirements]\n# Error scenarios must specify the exact HTTP status code returned\nrequire_http_status_code = true\n\n# Error scenarios must specify the exact error message or response body shape\nrequire_error_message = true\n\n# Error scenarios must state whether the node retries or fails immediately\nrequire_retry_behaviour = true\n\n[rules.input_boundary_requirements]\n# Every workflow that accepts string input MUST include these boundary scenarios\n[[rules.input_boundary_requirements.cases]]\ncase   = \"empty_string\"\ninput  = '\"\"'\nreason = \"empty input must have a defined rejection or fallback behaviour\"\n\n[[rules.input_boundary_requirements.cases]]\ncase   = \"whitespace_only\"\ninput  = '\"   \"'\nreason = \"whitespace-only must not be treated as valid content\"\n\n[[rules.input_boundary_requirements.cases]]\ncase   = \"null_or_missing\"\ninput  = \"null / field absent\"\nreason = \"missing input must return HTTP 400 with a specific error message\"\n\n[[rules.input_boundary_requirements.cases]]\ncase   = \"oversized_payload\"\ninput  = \"string exceeding max token or byte limit\"\nreason = \"must return HTTP 413 or equivalent with a size exceeded message\"\n\n[[rules.input_boundary_requirements.cases]]\ncase   = \"malformed_encoding\"\ninput  = \"non-UTF-8 or broken JSON\"\nreason = \"must return HTTP 400 with a parse error message\"\n\n[[rules.then_antipatterns]]\npattern = \"should generate an? (appropriate|correct|valid|concise) .*\"\nreason  = \"adjective without qualification is untestable \u2014 specify exact value or format\"\n\n[[rules.then_antipatterns]]\npattern = \"should handle .* appropriately\"\nreason  = \"unverifiable \u2014 specify the exact output or behaviour\"\n\n[[rules.then_antipatterns]]\npattern = \"should send an? (appropriate|correct|valid) response\"\nreason  = \"must specify the exact response value or structure\"\n\n[[rules.then_antipatterns]]\npattern = \"should fail gracefully\"\nreason  = \"specify the exact HTTP code, error message, and response shape\"\n\n[[rules.then_antipatterns]]\npattern = \"should return an error\"\nreason  = \"specify HTTP 4xx/5xx code, error field value, and whether retry is expected\"\n\n[[rules.then_antipatterns]]\npattern = \"should contain relevant information\"\nreason  = \"Verifier cannot match 'relevant' \u2014 specify exact field or substring\"\n\n[[rules.then_antipatterns]]\npattern = \"should respond within .* seconds\"\nreason  = \"Verifier is not a latency tester \u2014 remove timing assertions\"\n\n[[rules.then_antipatterns]]\npattern = \"should process the (input|message)\"\nreason  = \"processing is not observable \u2014 assert on the output only\"\n\n[input]\n# Receives the full requirements document from the Scribe\nsource   = \"scribe.output\"\ntemplate = \"{{ scribe.output }}\"\n\n[output]\n# Emit ONLY a single fenced code block \u2014 no preamble, no explanation\nformat = \"single fenced code block, label: behavior.feature\"\n\n[output.structure]\n# Feature name derived from workflow PURPOSE in the requirements doc\nfeature = \"<workflow name>\"\n\n# Required scenario types per processing step:\n# 1. happy path \u2014 concrete input, exact expected output\n# 2. error case \u2014 specific failure condition, HTTP code, exact error message\n# 3. input boundary \u2014 one scenario per boundary case for every string-accepting node\n[[output.structure.scenarios]]\ntype     = \"happy_path\"\ntemplate = \"\"\"\nScenario: <happy path name>\n  Given a chat message \"<exact literal input string>\"\n  When <node_id> processes the message\n  Then <node_id> should return \"<exact expected output string>\"\n  And reply_chat should send \"<exact message>\" back to the chat caller\n\"\"\"\n\n[[output.structure.scenarios]]\ntype     = \"error_case\"\ntemplate = \"\"\"\nScenario: <error case name>\n  Given a chat message \"<exact literal input string>\"\n  When <node_id> encounters <specific failure condition>\n  Then <node_id> should return HTTP <status_code> with error \"<exact error message>\"\n  And reply_chat should send \"<exact fallback message>\" back to the chat caller\n\"\"\"\n\n[[output.structure.scenarios]]\ntype     = \"input_boundary\"\ntemplate = \"\"\"\nScenario: <boundary case name>\n  Given a <empty|whitespace|null|oversized|malformed> chat message <exact input value>\n  When <node_id> attempts to process the input\n  Then <node_id> should return HTTP <status_code> with error \"<exact error message>\"\n  And reply_chat should send \"<exact response>\" back to the chat caller\n\"\"\"\n\n[validation]\n# Before returning your output, you MUST call the validate_gherkin tool\n# to check your Gherkin spec for syntax errors and lint warnings.\n# Fix any issues the tool reports before returning your final output.\nrequire_validation = true\ntool_name = \"validate_gherkin\"\n\n[validation.workflow]\n# 1. Generate your Gherkin spec\n# 2. Call validate_gherkin with the raw spec text (no fences)\n# 3. If valid=false or lint_errors exist, fix the issues and re-validate\n# 4. Only return the spec once it passes validation\nmust_pass_before_output = true\n",
                 "extra_config": {
                     "input_template": "{{ scribe.output }}",
                     "conversation_memory": false,
@@ -221,6 +221,68 @@
             "updated_at": "2026-03-20T09:05:20",
             "schedule_job": null
         },
+        {
+            "id": 27,
+            "node_id": "gherkin_check",
+            "label": "Gherkin Valid?",
+            "component_type": "assertion",
+            "is_entry_point": false,
+            "interrupt_before": false,
+            "interrupt_after": false,
+            "position_x": 700,
+            "position_y": -100,
+            "config": {
+                "system_prompt": "",
+                "extra_config": {
+                    "rules": [
+                        {
+                            "id": "has_feature",
+                            "field": "node_outputs.gherkin_agent.output",
+                            "operator": "contains",
+                            "value": "Feature:"
+                        },
+                        {
+                            "id": "has_scenario",
+                            "field": "node_outputs.gherkin_agent.output",
+                            "operator": "contains",
+                            "value": "Scenario:"
+                        },
+                        {
+                            "id": "has_given",
+                            "field": "node_outputs.gherkin_agent.output",
+                            "operator": "contains",
+                            "value": "Given"
+                        },
+                        {
+                            "id": "has_then",
+                            "field": "node_outputs.gherkin_agent.output",
+                            "operator": "contains",
+                            "value": "Then"
+                        }
+                    ]
+                },
+                "llm_credential_id": null,
+                "model_name": "",
+                "temperature": null,
+                "max_tokens": null,
+                "frequency_penalty": null,
+                "presence_penalty": null,
+                "top_p": null,
+                "timeout": null,
+                "max_retries": null,
+                "response_format": null,
+                "llm_model_config_id": null,
+                "credential_id": null,
+                "is_active": true,
+                "priority": 0,
+                "trigger_config": {},
+                "input_template": null
+            },
+            "subworkflow_id": null,
+            "code_block_id": null,
+            "updated_at": "2026-03-21T07:31:26",
+            "schedule_job": null
+        },
         {
             "id": 12,
             "node_id": "gherkin_model",
@@ -229,8 +291,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 113,
-            "position_y": -2,
+            "position_x": 83,
+            "position_y": -85,
             "config": {
                 "system_prompt": "",
                 "extra_config": {},
@@ -253,7 +315,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T13:05:32",
+            "updated_at": "2026-03-21T06:08:37",
             "schedule_job": null
         },
         {
@@ -264,8 +326,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 517,
-            "position_y": -48,
+            "position_x": 549,
+            "position_y": -42,
             "config": {
                 "system_prompt": "",
                 "extra_config": {
@@ -294,7 +356,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T09:01:27",
+            "updated_at": "2026-03-20T21:59:55",
             "schedule_job": null
         },
         {
@@ -305,8 +367,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": -116,
-            "position_y": 151,
+            "position_x": -98,
+            "position_y": 101,
             "config": {
                 "system_prompt": "{{ scribe.output }}",
                 "extra_config": {},
@@ -329,7 +391,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T09:01:27",
+            "updated_at": "2026-03-21T06:08:33",
             "schedule_job": null
         },
         {
@@ -375,10 +437,10 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 1482,
-            "position_y": -141,
+            "position_x": 1480,
+            "position_y": -165,
             "config": {
-                "system_prompt": "✅ Verification passed! Here is what I designed:\n\n**Topology:**\n{{ topology_agent.output }}\n\n---\n\n**Behavior Spec:**\n{{ gherkin_agent.output }}",
+                "system_prompt": "\u2705 Verification passed! Here is what I designed:\n\n**Topology:**\n{{ topology_agent.output }}\n\n---\n\n**Behavior Spec:**\n{{ gherkin_agent.output }}",
                 "extra_config": {},
                 "llm_credential_id": null,
                 "model_name": "",
@@ -399,7 +461,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T09:01:27",
+            "updated_at": "2026-03-20T21:52:11",
             "schedule_job": null
         },
         {
@@ -413,7 +475,7 @@
             "position_x": -840,
             "position_y": -45,
             "config": {
-                "system_prompt": "[agent]\nname = \"Scribe\"\nrole = \"\"\"\nYou are the Scribe. Your job is to turn the user's request into a structured requirements document that feeds a validation gate, a Gherkin test scenario agent, and a topology/DSL builder agent.\n\"\"\"\n\n[rules]\n# Always produce the requirements document immediately, even if the request is incomplete\nalways_produce_document = true\n\n# Do not ask the user directly — the validation layer owns the clarification loop\nask_user_directly = false\nclarification_owner = \"validation layer\"\n\n# The trigger (chat, schedule, manual, telegram) IS the entry point — do NOT list a separate \"receive\" or \"capture\" step\ntrigger_is_entry_point = true\n\n# STEPS should only include PROCESSING nodes\nsteps_processing_only = true\n\n# Make reasonable assumptions for anything unspecified — record them in ASSUMPTIONS\n# If anything is genuinely unresolvable without user input, record it in OPEN_QUESTIONS\nrecord_assumptions = true\nrecord_open_questions = true\n\n[node_catalog]\n\n# Triggers (entry points — every workflow needs exactly one)\ntrigger_chat              = \"Receives messages from external chat clients via msg-gateway generic adapter\"\ntrigger_error             = \"Triggered when a workflow execution encounters an error\"\ntrigger_manual            = \"Manually triggered workflow execution\"\ntrigger_schedule          = \"Executes workflow on a scheduled interval\"\ntrigger_telegram          = \"Receives messages from Telegram via msg-gateway\"\ntrigger_workflow          = \"Triggered by another workflow execution\"\n\n# AI Nodes (call an LLM to reason, classify, generate, extract)\nagent                     = \"LangGraph react agent with tools\"\ncategorizer               = \"Classifies input into categories\"\ndeep_agent                = \"Advanced agent with built-in task planning, filesystem tools, and subagents\"\nextractor                 = \"Extracts structured data from input\"\nrouter                    = \"Routes to different branches based on input\"\n\n# Logic Nodes (no LLM — deterministic control flow)\ncode                      = \"Python sandbox for data transformation, API calls, parsing\"\nfilter                    = \"Filter array items using rule-based matching\"\nloop                      = \"Iterate over an array, executing body nodes for each item\"\nmerge                     = \"Merge outputs from multiple branches into one (fan-in barrier)\"\nswitch                    = \"Routes to different branches based on a state field or expression\"\nwait                      = \"Delay downstream execution by a specified duration\"\nworkflow                  = \"Execute another workflow as a child and return its output\"\n\n# Flow Control\nhuman_confirmation        = \"Pause execution and wait for human approval before continuing\"\n\n# Output Nodes (terminal — end the workflow)\nreply_chat                = \"Sends a message back to the chat caller and ends the workflow\"\n\n# security and privileged access\nidentify_user             = \"Only relevant as a step before totp, which means whenever there is intention to acquire privileged access to information or actions\"\n\n# Agent Tools (attach to agent/deep_agent via tool edges)\nget_totp_code             = \"Retrieve the current TOTP code for agent identity verification\"\nplatform_api              = \"Make authenticated requests to the platform API\"\nscheduler_tools           = \"Create, pause, resume, stop, and list scheduled recurring jobs\"\nspawn_and_await           = \"Spawn a child workflow and wait for its result inside an agent's reasoning loop\"\nsystem_health             = \"Check platform infrastructure health: Redis, RQ workers, queues, stuck executions\"\nwhoami                    = \"Get self-awareness — workflow, node ID, and how to modify yourself\"\nworkflow_create           = \"Create workflows programmatically from a YAML DSL specification\"\nworkflow_discover         = \"Search existing workflows by requirements and get reuse recommendations\"\n\n# Sub-components (wired by the builder, not listed in STEPS)\nai_model                  = \"LLM model configuration — attached to agent/deep_agent via llm edge\"\nmemory_read               = \"Recall tool — retrieves information from global memory\"\nmemory_write              = \"Remember tool — stores information in global memory\"\noutput_parser             = \"Structured output parsing for agent responses\"\nrun_command               = \"Execute shell commands\"\nskill                     = \"SKILL.md behavioral instructions for agents via progressive disclosure\"\n\n[output.requirements]\nformat = \"fenced code block, label: requirements\"\nfields = [\n  \"PURPOSE: <one sentence>\",\n  \"TRIGGER: <chat|schedule|manual|telegram>\",\n  \"STEPS: <numbered list of node_id: description [type]>\",\n  \"INTEGRATIONS: <list or none>\",\n  \"ASSUMPTIONS: <list of decisions the Scribe made on behalf of the user>\",\n  \"OPEN_QUESTIONS: <unresolved questions requiring user clarification — empty list if none>\",\n  \"ERROR_HANDLING: <strategy>\",\n  \"OUTPUT: <what the workflow produces>\",\n]\n\n[output.mermaid]\nformat = \"fenced code block, label: mermaid\"\ndiagram_type = \"flowchart TD\"\n\n[output.mermaid.rules]\nregular_nodes  = \"Square brackets []\"\nswitch_nodes   = \"Diamond shapes {}\"\nterminal_nodes = \"Stadium shape (( ))\"\nbranch_labels  = true\nparallel_paths = true\n\n[output.order]\n1 = \"requirements block\"\n2 = \"mermaid block\"\n3 = \"closing prompt: 'Does this match what you had in mind? I can adjust before proceeding.'\"\n\n[downstream]\nvalidator    = \"Gates on OPEN_QUESTIONS — empty means pass, non-empty triggers clarification back to user\"\ngherkin      = \"Generates test scenario scripts from full requirements document\"\ntopology_dsl = \"Builds workflow DSL from full requirements document\"\n",
+                "system_prompt": "[agent]\nname = \"Scribe\"\nrole = \"\"\"\nYou are the Scribe. Your job is to turn the user's request into a structured requirements document that feeds a validation gate, a Gherkin test scenario agent, and a topology/DSL builder agent.\n\"\"\"\n\n[rules]\n# Always produce the requirements document immediately, even if the request is incomplete\nalways_produce_document = true\n\n# Do not ask the user directly \u2014 the validation layer owns the clarification loop\nask_user_directly = false\nclarification_owner = \"validation layer\"\n\n# The trigger (chat, schedule, manual, telegram) IS the entry point \u2014 do NOT list a separate \"receive\" or \"capture\" step\ntrigger_is_entry_point = true\n\n# STEPS should only include PROCESSING nodes\nsteps_processing_only = true\n\n# Make reasonable assumptions for anything unspecified \u2014 record them in ASSUMPTIONS\n# If anything is genuinely unresolvable without user input, record it in OPEN_QUESTIONS\nrecord_assumptions = true\nrecord_open_questions = true\n\n[node_catalog]\n\n# Triggers (entry points \u2014 every workflow needs exactly one)\ntrigger_chat              = \"Receives messages from external chat clients via msg-gateway generic adapter\"\ntrigger_error             = \"Triggered when a workflow execution encounters an error\"\ntrigger_manual            = \"Manually triggered workflow execution\"\ntrigger_schedule          = \"Executes workflow on a scheduled interval\"\ntrigger_telegram          = \"Receives messages from Telegram via msg-gateway\"\ntrigger_workflow          = \"Triggered by another workflow execution\"\n\n# AI Nodes (call an LLM to reason, classify, generate, extract)\nagent                     = \"LangGraph react agent with tools\"\ncategorizer               = \"Classifies input into categories\"\ndeep_agent                = \"Advanced agent with built-in task planning, filesystem tools, and subagents\"\nextractor                 = \"Extracts structured data from input\"\nrouter                    = \"Routes to different branches based on input\"\n\n# Logic Nodes (no LLM \u2014 deterministic control flow)\ncode                      = \"Python sandbox for data transformation, API calls, parsing\"\nfilter                    = \"Filter array items using rule-based matching\"\nloop                      = \"Iterate over an array, executing body nodes for each item\"\nmerge                     = \"Merge outputs from multiple branches into one (fan-in barrier)\"\nswitch                    = \"Routes to different branches based on a state field or expression\"\nwait                      = \"Delay downstream execution by a specified duration\"\nworkflow                  = \"Execute another workflow as a child and return its output\"\n\n# Flow Control\nhuman_confirmation        = \"Pause execution and wait for human approval before continuing\"\n\n# Output Nodes (terminal \u2014 end the workflow)\nreply_chat                = \"Sends a message back to the chat caller and ends the workflow\"\n\n# security and privileged access\nidentify_user             = \"Only relevant as a step before totp, which means whenever there is intention to acquire privileged access to information or actions\"\n\n# Agent Tools (attach to agent/deep_agent via tool edges)\nepic_tools                = \"Create, query, update, and search epics for task delegation\"\nget_totp_code             = \"Retrieve the current TOTP code for agent identity verification\"\nplatform_api              = \"Make authenticated requests to the platform API\"\nscheduler_tools           = \"Create, pause, resume, stop, and list scheduled recurring jobs\"\nspawn_and_await           = \"Spawn a child workflow and wait for its result inside an agent's reasoning loop\"\nsystem_health             = \"Check platform infrastructure health: Redis, RQ workers, queues, stuck executions\"\ntask_tools                = \"Create, list, update, and cancel tasks within epics\"\nwhoami                    = \"Get self-awareness \u2014 workflow, node ID, and how to modify yourself\"\nworkflow_create           = \"Create workflows programmatically from a YAML DSL specification\"\nworkflow_discover         = \"Search existing workflows by requirements and get reuse recommendations\"\n\n# Sub-components (wired by the builder, not listed in STEPS)\nai_model                  = \"LLM model configuration \u2014 attached to agent/deep_agent via llm edge\"\nmemory_read               = \"Recall tool \u2014 retrieves information from global memory\"\nmemory_write              = \"Remember tool \u2014 stores information in global memory\"\noutput_parser             = \"Structured output parsing for agent responses\"\nrun_command               = \"Execute shell commands\"\nskill                     = \"SKILL.md behavioral instructions for agents via progressive disclosure\"\n\n[output.requirements]\nformat = \"fenced code block, label: requirements\"\nfields = [\n  \"PURPOSE: <one sentence>\",\n  \"TRIGGER: <chat|schedule|manual|telegram>\",\n  \"STEPS: <numbered list of node_id: description [type]>\",\n  \"INTEGRATIONS: <list or none>\",\n  \"ASSUMPTIONS: <list of decisions the Scribe made on behalf of the user>\",\n  \"OPEN_QUESTIONS: <unresolved questions requiring user clarification \u2014 empty list if none>\",\n  \"ERROR_HANDLING: <strategy>\",\n  \"OUTPUT: <what the workflow produces>\",\n]\n\n[output.mermaid]\nformat = \"fenced code block, label: mermaid\"\ndiagram_type = \"flowchart TD\"\n\n[output.mermaid.rules]\nregular_nodes  = \"Square brackets []\"\nswitch_nodes   = \"Diamond shapes {}\"\nterminal_nodes = \"Stadium shape (( ))\"\nbranch_labels  = true\nparallel_paths = true\n\n[output.order]\n1 = \"requirements block\"\n2 = \"mermaid block\"\n3 = \"closing prompt: 'Does this match what you had in mind? I can adjust before proceeding.'\"\n\n[downstream]\nvalidator    = \"Gates on OPEN_QUESTIONS \u2014 empty means pass, non-empty triggers clarification back to user\"\ngherkin      = \"Generates test scenario scripts from full requirements document\"\ntopology_dsl = \"Builds workflow DSL from full requirements document\"\n",
                 "extra_config": {
                     "conversation_memory": false,
                     "context_window": null,
@@ -499,8 +561,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": -995,
-            "position_y": 178,
+            "position_x": -940,
+            "position_y": 134,
             "config": {
                 "system_prompt": "",
                 "extra_config": {},
@@ -523,7 +585,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T09:01:27",
+            "updated_at": "2026-03-20T22:02:23",
             "schedule_job": null
         },
         {
@@ -534,10 +596,10 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 178,
-            "position_y": 90,
+            "position_x": 186,
+            "position_y": 61,
             "config": {
-                "system_prompt": "[agent]\nname = \"Topology Agent\"\nrole = \"You are the Topology Agent. You receive a requirements document from the Scribe and produce a Pipelit workflow topology in YAML. Your output is the structural blueprint -- node types, IDs, and edges only. Prompts, code, and runtime config are filled in later by the Builder.\"\n\n[context]\n# The trigger IS the entry point -- do NOT create a separate receive/capture node\ntrigger_is_entry_point = true\n\n# Topology is STRUCTURE ONLY -- no prompts, no code snippets, no config values\nstructure_only = true\n\n# Use the exact node IDs from the Scribe's STEPS section\nuse_exact_node_ids = true\n\n[node_catalog]\n\n# Triggers (entry points -- every workflow needs exactly one)\ntrigger_chat              = \"Receives messages from external chat clients\"\ntrigger_error             = \"Triggered when a workflow execution encounters an error\"\ntrigger_manual            = \"Manually triggered workflow execution\"\ntrigger_schedule          = \"Executes workflow on a scheduled interval\"\ntrigger_telegram          = \"Receives messages from Telegram via msg-gateway\"\ntrigger_workflow          = \"Triggered by another workflow execution\"\n\n# AI Nodes (call an LLM)\nagent                     = \"LangGraph react agent with tools\"\ncategorizer               = \"Classifies input into categories\"\ndeep_agent                = \"Advanced agent with planning, filesystem tools, and subagents\"\nextractor                 = \"Extracts structured data from input\"\nrouter                    = \"Routes to different branches based on LLM classification\"\n\n# Logic Nodes (no LLM -- deterministic)\ncode                      = \"Python sandbox for data transformation, API calls, parsing\"\nfilter                    = \"Filter array items using rule-based matching\"\nloop                      = \"Iterate over an array, executing body nodes for each item\"\nmerge                     = \"Fan-in barrier -- waits for all incoming parallel branches\"\nswitch                    = \"Routes to branches based on a state field or expression\"\nwait                      = \"Delay downstream execution by a specified duration\"\nworkflow                  = \"Execute another workflow as a child and return its output\"\n\n# Flow Control\nhuman_confirmation        = \"Pause execution and wait for human approval\"\n\n# Output Nodes (terminal -- end the workflow)\nreply_chat                = \"Sends a message back to the chat caller and ends the workflow\"\nerror_handler             = \"Catches errors from upstream nodes\"\n\n# Agent Tools (attach via tool edges -- not listed in steps)\nplatform_api              = \"Make authenticated requests to the platform API\"\nworkflow_create           = \"Create workflows from YAML DSL\"\nspawn_and_await           = \"Spawn a child workflow and wait for its result\"\nscheduler_tools           = \"Manage scheduled recurring jobs\"\n\n# Sub-components (wired by the builder)\nai_model                  = \"LLM model config -- attached to agent/deep_agent via llm edge\"\nmemory_read               = \"Recall tool -- retrieves from global memory\"\nmemory_write              = \"Remember tool -- stores to global memory\"\nskill                     = \"SKILL.md behavioral instructions for agents\"\n\n[rules]\n# Do NOT include trigger as a step -- it goes in the trigger: field only\ntrigger_not_in_steps = true\n\n# Every step from requirements maps to exactly one node\none_node_per_step = true\n\n# Use snake_case for all node IDs\nsnake_case_ids = true\n\n# For non-linear flows, include explicit edge definitions\nexplicit_edges_for_branches = true\n\n[rules.edges]\n# Default flow is linear top-to-bottom (step 1 -> step 2 -> step 3)\n# Only specify edges explicitly when flow is non-linear:\n# - switch branches (conditional edges with labels)\n# - parallel fan-out (one node -> multiple nodes)\n# - loops (back-edges)\n# - merge (multiple nodes -> one node)\nonly_explicit_for_nonlinear = true\n\n[input]\nsource   = \"scribe.output\"\ntemplate = \"{{ scribe.output }}\"\n\n[output]\nformat = \"single fenced code block, label: topology.yaml\"\n# No preamble, no explanation -- ONLY the fenced code block\n\n[output.yaml_structure]\n# Top-level fields\nname    = \"<workflow name from requirements PURPOSE>\"\nslug    = \"<kebab-case-slug>\"\ntrigger = \"<trigger type from requirements TRIGGER>\"\n\n[output.yaml_structure.model]\ncredential_id = 1\nname          = \"claude-sonnet-4-20250514\"\n\n[output.yaml_structure.steps]\n# Ordered list matching the Scribe's STEPS\n# Each step has: type, id, label\ntype  = \"<component_type from node_catalog>\"\nid    = \"<exact node_id from requirements STEPS>\"\nlabel = \"<human-readable display name>\"\n\n[output.yaml_structure.edges]\n# Only include when flow is non-linear\nfrom      = \"<source_node_id>\"\nto        = \"<target_node_id>\"\ncondition = \"<label -- only for switch branches>\"\n\n[validation]\n# Before returning your output, you MUST call the validate_topology tool\n# to check your YAML topology for structural errors.\n# Fix any issues the tool reports before returning your final output.\nrequire_validation = true\ntool_name = \"validate_topology\"\n\n[validation.workflow]\n# 1. Generate your topology YAML\n# 2. Call validate_topology with the raw YAML text (no fences)\n# 3. If valid=false or errors exist, fix the issues and re-validate\n# 4. Only return the topology once it passes validation\nmust_pass_before_output = true\n",
+                "system_prompt": "[agent]\nname = \"Topology Agent\"\nrole = \"You are the Topology Agent. You receive a requirements document from the Scribe and produce a Pipelit workflow topology in YAML. Your output is the structural blueprint -- node types, IDs, and edges only. Prompts, code, and runtime config are filled in later by the Builder.\"\n\n[context]\n# The trigger IS the entry point -- do NOT create a separate receive/capture node\ntrigger_is_entry_point = true\n\n# Topology is STRUCTURE ONLY -- no prompts, no code snippets, no config values\nstructure_only = true\n\n# Use the exact node IDs from the Scribe's STEPS section\nuse_exact_node_ids = true\n\n[node_catalog]\n\n# Triggers (entry points -- every workflow needs exactly one)\ntrigger_chat              = \"Receives messages from external chat clients\"\ntrigger_error             = \"Triggered when a workflow execution encounters an error\"\ntrigger_manual            = \"Manually triggered workflow execution\"\ntrigger_schedule          = \"Executes workflow on a scheduled interval\"\ntrigger_telegram          = \"Receives messages from Telegram via msg-gateway\"\ntrigger_workflow          = \"Triggered by another workflow execution\"\n\n# AI Nodes (call an LLM)\nagent                     = \"LangGraph react agent with tools\"\ncategorizer               = \"Classifies input into categories\"\ndeep_agent                = \"Advanced agent with planning, filesystem tools, and subagents\"\nextractor                 = \"Extracts structured data from input\"\nrouter                    = \"Routes to different branches based on LLM classification\"\n\n# Logic Nodes (no LLM -- deterministic)\ncode                      = \"Python sandbox for data transformation, API calls, parsing\"\nfilter                    = \"Filter array items using rule-based matching\"\nloop                      = \"Iterate over an array, executing body nodes for each item\"\nmerge                     = \"Fan-in barrier -- waits for all incoming parallel branches\"\nswitch                    = \"Routes to branches based on a state field or expression\"\nwait                      = \"Delay downstream execution by a specified duration\"\nworkflow                  = \"Execute another workflow as a child and return its output\"\n\n# Flow Control\nhuman_confirmation        = \"Pause execution and wait for human approval\"\n\n# Output Nodes (terminal -- end the workflow)\nreply_chat                = \"Sends a message back to the chat caller and ends the workflow\"\nerror_handler             = \"Catches errors from upstream nodes\"\n\n# Agent Tools (attach via tool edges -- not listed in steps)\nplatform_api              = \"Make authenticated requests to the platform API\"\nworkflow_create           = \"Create workflows from YAML DSL\"\nspawn_and_await           = \"Spawn a child workflow and wait for its result\"\nepic_tools                = \"Create, query, update epics\"\ntask_tools                = \"Create, list, update, cancel tasks within epics\"\nscheduler_tools           = \"Manage scheduled recurring jobs\"\n\n# Sub-components (wired by the builder)\nai_model                  = \"LLM model config -- attached to agent/deep_agent via llm edge\"\nmemory_read               = \"Recall tool -- retrieves from global memory\"\nmemory_write              = \"Remember tool -- stores to global memory\"\nskill                     = \"SKILL.md behavioral instructions for agents\"\n\n[rules]\n# Do NOT include trigger as a step -- it goes in the trigger: field only\ntrigger_not_in_steps = true\n\n# Every step from requirements maps to exactly one node\none_node_per_step = true\n\n# Use snake_case for all node IDs\nsnake_case_ids = true\n\n# For non-linear flows, include explicit edge definitions\nexplicit_edges_for_branches = true\n\n[rules.edges]\n# Default flow is linear top-to-bottom (step 1 -> step 2 -> step 3)\n# Only specify edges explicitly when flow is non-linear:\n# - switch branches (conditional edges with labels)\n# - parallel fan-out (one node -> multiple nodes)\n# - loops (back-edges)\n# - merge (multiple nodes -> one node)\nonly_explicit_for_nonlinear = true\n\n[input]\nsource   = \"scribe.output\"\ntemplate = \"{{ scribe.output }}\"\n\n[output]\nformat = \"single fenced code block, label: topology.yaml\"\n# No preamble, no explanation -- ONLY the fenced code block\n\n[output.yaml_structure]\n# Top-level fields\nname    = \"<workflow name from requirements PURPOSE>\"\nslug    = \"<kebab-case-slug>\"\ntrigger = \"<trigger type from requirements TRIGGER>\"\n\n[output.yaml_structure.model]\ncredential_id = 1\nname          = \"claude-sonnet-4-20250514\"\n\n[output.yaml_structure.steps]\n# Ordered list matching the Scribe's STEPS\n# Each step has: type, id, label\ntype  = \"<component_type from node_catalog>\"\nid    = \"<exact node_id from requirements STEPS>\"\nlabel = \"<human-readable display name>\"\n\n[output.yaml_structure.edges]\n# Only include when flow is non-linear\nfrom      = \"<source_node_id>\"\nto        = \"<target_node_id>\"\ncondition = \"<label -- only for switch branches>\"\n\n[validation]\n# Before returning your output, you MUST call the validate_topology tool\n# to check your YAML topology for structural errors.\n# Fix any issues the tool reports before returning your final output.\nrequire_validation = true\ntool_name = \"validate_topology\"\n\n[validation.workflow]\n# 1. Generate your topology YAML\n# 2. Call validate_topology with the raw YAML text (no fences)\n# 3. If valid=false or errors exist, fix the issues and re-validate\n# 4. Only return the topology once it passes validation\nmust_pass_before_output = true\n",
                 "extra_config": {
                     "input_template": "{{ scribe.output }}"
                 },
@@ -560,7 +622,69 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T09:01:27",
+            "updated_at": "2026-03-21T06:08:48",
+            "schedule_job": null
+        },
+        {
+            "id": 28,
+            "node_id": "topology_check",
+            "label": "Topology Valid?",
+            "component_type": "assertion",
+            "is_entry_point": false,
+            "interrupt_before": false,
+            "interrupt_after": false,
+            "position_x": 700,
+            "position_y": 100,
+            "config": {
+                "system_prompt": "",
+                "extra_config": {
+                    "rules": [
+                        {
+                            "id": "has_name",
+                            "field": "node_outputs.topology_agent.output",
+                            "operator": "contains",
+                            "value": "name:"
+                        },
+                        {
+                            "id": "has_trigger",
+                            "field": "node_outputs.topology_agent.output",
+                            "operator": "contains",
+                            "value": "trigger:"
+                        },
+                        {
+                            "id": "has_steps",
+                            "field": "node_outputs.topology_agent.output",
+                            "operator": "contains",
+                            "value": "steps:"
+                        },
+                        {
+                            "id": "has_type",
+                            "field": "node_outputs.topology_agent.output",
+                            "operator": "contains",
+                            "value": "type:"
+                        }
+                    ]
+                },
+                "llm_credential_id": null,
+                "model_name": "",
+                "temperature": null,
+                "max_tokens": null,
+                "frequency_penalty": null,
+                "presence_penalty": null,
+                "top_p": null,
+                "timeout": null,
+                "max_retries": null,
+                "response_format": null,
+                "llm_model_config_id": null,
+                "credential_id": null,
+                "is_active": true,
+                "priority": 0,
+                "trigger_config": {},
+                "input_template": null
+            },
+            "subworkflow_id": null,
+            "code_block_id": null,
+            "updated_at": "2026-03-21T07:31:26",
             "schedule_job": null
         },
         {
@@ -571,8 +695,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 75,
-            "position_y": 341,
+            "position_x": 35,
+            "position_y": 251,
             "config": {
                 "system_prompt": "",
                 "extra_config": {},
@@ -595,7 +719,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T13:05:31",
+            "updated_at": "2026-03-21T06:09:01",
             "schedule_job": null
         },
         {
@@ -641,8 +765,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 311,
-            "position_y": -80,
+            "position_x": 335,
+            "position_y": -75,
             "config": {
                 "system_prompt": "",
                 "extra_config": {},
@@ -665,7 +789,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T13:05:34",
+            "updated_at": "2026-03-21T06:08:36",
             "schedule_job": null
         },
         {
@@ -676,8 +800,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 351,
-            "position_y": 296,
+            "position_x": 317,
+            "position_y": 249,
             "config": {
                 "system_prompt": "",
                 "extra_config": {},
@@ -700,7 +824,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T13:05:31",
+            "updated_at": "2026-03-21T06:09:02",
             "schedule_job": null
         },
         {
@@ -711,8 +835,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 748,
-            "position_y": -20,
+            "position_x": 766,
+            "position_y": -67,
             "config": {
                 "system_prompt": "[agent]\nname = \"Verifier\"\nrole = \"You are the Verifier. You receive a Gherkin behavior spec and a topology YAML (produced independently by different agents from the same requirements) and cross-check them for consistency.\"\n\n[input]\ngherkin_spec = \"{{ gherkin_agent.output }}\"\ntopology_yaml = \"{{ topology_agent.output }}\"\n\n[checks]\n\n[checks.coverage]\ndescription = \"Every node in the topology steps must appear in at least one Gherkin scenario\"\ndirection   = \"bidirectional\"\n# Also check: every node referenced in Gherkin must exist in topology steps\n# Trigger node is implicit -- do NOT require it in Gherkin scenarios\n\n[checks.types]\ndescription = \"Node types in topology must match how scenarios describe each node's behaviour\"\n# e.g. if topology says type: agent, the scenario should describe LLM reasoning, not code execution\n\n[checks.flow]\ndescription = \"The edge flow in topology must support the Given/When/Then sequence in scenarios\"\n# Steps execute top-to-bottom by default (step 1 -> step 2 -> step 3)\n# The edges section is ONLY for non-linear flow: fan-out, fan-in, switch branches, loops\n# If steps are [A, B, C] and edges only define fan-out/switch, the implicit linear\n# connections A->B->C still exist -- do NOT flag missing edges for linear sequences\nlinear_flow_is_implicit = true\n\n[checks.completeness]\ndescription = \"Error scenarios and input boundary scenarios must exist\"\n# At minimum: one happy path, one error case, one boundary case per string-accepting node\n\n[checks.id_consistency]\ndescription = \"Node IDs must match exactly between topology and Gherkin\"\n# Both agents receive the same Scribe output with canonical node IDs\n# If IDs diverge, this is a fail\n\n[rules]\n# Do NOT invent or hallucinate nodes, scenarios, or content that isn't in the input\nno_hallucination = true\n\n# Do NOT verify trigger_chat/trigger_* in Gherkin -- triggers are implicit infrastructure\ntrigger_is_implicit = true\n\n# Be strict on ID matching but lenient on label/display name differences\nstrict_id_matching = true\nlenient_label_matching = true\n\n[output]\n# Output ONLY a JSON object -- no markdown fencing, no explanation, no preamble\nformat = \"raw JSON only\"\n\n[output.pass]\ntemplate = '{\"verdict\": \"pass\", \"issues\": []}'\n\n[output.fail]\ntemplate = '{\"verdict\": \"fail\", \"issues\": [\"issue 1 description\", \"issue 2 description\"]}'\n\n[output.rules]\n# Each issue must be specific and actionable\n# Bad: \"coverage issue\" -- Good: \"node 'summarizer' in topology not referenced in any Gherkin scenario\"\nissues_must_be_specific = true\n\n# Maximum 5 issues -- prioritize the most critical\nmax_issues = 5\n",
                 "extra_config": {},
@@ -735,7 +859,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T09:01:27",
+            "updated_at": "2026-03-20T21:59:56",
             "schedule_job": null
         },
         {
@@ -746,8 +870,8 @@
             "is_entry_point": false,
             "interrupt_before": false,
             "interrupt_after": false,
-            "position_x": 631,
-            "position_y": 156,
+            "position_x": 709,
+            "position_y": 103,
             "config": {
                 "system_prompt": "",
                 "extra_config": {},
@@ -770,7 +894,7 @@
             },
             "subworkflow_id": null,
             "code_block_id": null,
-            "updated_at": "2026-03-20T13:09:34",
+            "updated_at": "2026-03-20T22:09:13",
             "schedule_job": null
         }
     ],
@@ -964,6 +1088,26 @@
             "condition_mapping": null,
             "condition_value": "",
             "priority": 0
+        },
+        {
+            "id": 25,
+            "source_node_id": "gherkin_agent",
+            "target_node_id": "gherkin_check",
+            "edge_type": "direct",
+            "edge_label": "",
+            "condition_mapping": null,
+            "condition_value": "",
+            "priority": 0
+        },
+        {
+            "id": 26,
+            "source_node_id": "topology_agent",
+            "target_node_id": "topology_check",
+            "edge_type": "direct",
+            "edge_label": "",
+            "condition_mapping": null,
+            "condition_value": "",
+            "priority": 0
         }
     ]
 }
diff --git a/platform/tests/test_assertion.py b/platform/tests/test_assertion.py
new file mode 100644
index 0000000..fbc1852
--- /dev/null
+++ b/platform/tests/test_assertion.py
@@ -0,0 +1,121 @@
+"""Tests for assertion component — rule evaluation with pass/fail routing."""
+
+from __future__ import annotations
+
+from unittest.mock import MagicMock
+
+from components.assertion import assertion_factory
+
+
+def _make_node(extra_config: dict):
+    """Build a mock node with the given extra_config."""
+    node = MagicMock()
+    node.component_config.extra_config = extra_config
+    return node
+
+
+class TestAssertionComponent:
+    def test_all_rules_pass(self):
+        node = _make_node({
+            "rules": [
+                {"id": "r1", "field": "status", "operator": "equals", "value": "active"},
+                {"id": "r2", "field": "count", "operator": "gt", "value": "0"},
+            ],
+        })
+        fn = assertion_factory(node)
+        result = fn({"status": "active", "count": 5})
+
+        assert result["_route"] == "pass"
+        assert result["output"]["passed"] is True
+        assert len(result["output"]["results"]) == 2
+        assert all(r["passed"] for r in result["output"]["results"])
+
+    def test_one_rule_fails(self):
+        node = _make_node({
+            "rules": [
+                {"id": "r1", "field": "status", "operator": "equals", "value": "active"},
+                {"id": "r2", "field": "status", "operator": "equals", "value": "deleted"},
+            ],
+        })
+        fn = assertion_factory(node)
+        result = fn({"status": "active"})
+
+        assert result["_route"] == "fail"
+        assert result["output"]["passed"] is False
+        assert result["output"]["results"][0]["passed"] is True
+        assert result["output"]["results"][1]["passed"] is False
+
+    def test_multiple_rules_mixed_results(self):
+        node = _make_node({
+            "rules": [
+                {"id": "r1", "field": "name", "operator": "contains", "value": "foo"},
+                {"id": "r2", "field": "age", "operator": "gt", "value": "18"},
+                {"id": "r3", "field": "role", "operator": "equals", "value": "admin"},
+            ],
+        })
+        fn = assertion_factory(node)
+        result = fn({"name": "foobar", "age": 10, "role": "admin"})
+
+        assert result["_route"] == "fail"
+        assert result["output"]["passed"] is False
+        results = result["output"]["results"]
+        assert results[0]["passed"] is True   # name contains foo
+        assert results[1]["passed"] is False  # age not > 18
+        assert results[2]["passed"] is True   # role equals admin
+
+    def test_empty_rules_passes(self):
+        node = _make_node({"rules": []})
+        fn = assertion_factory(node)
+        result = fn({"anything": "value"})
+
+        assert result["_route"] == "pass"
+        assert result["output"]["passed"] is True
+        assert result["output"]["results"] == []
+
+    def test_no_rules_key_passes(self):
+        node = _make_node({})
+        fn = assertion_factory(node)
+        result = fn({"anything": "value"})
+
+        assert result["_route"] == "pass"
+        assert result["output"]["passed"] is True
+
+    def test_missing_field_fails_gracefully(self):
+        node = _make_node({
+            "rules": [
+                {"id": "r1", "field": "nonexistent", "operator": "equals", "value": "something"},
+            ],
+        })
+        fn = assertion_factory(node)
+        result = fn({"other_field": "value"})
+
+        assert result["_route"] == "fail"
+        assert result["output"]["passed"] is False
+        assert result["output"]["results"][0]["passed"] is False
+        assert result["output"]["results"][0]["actual"] is None
+
+    def test_nested_field_resolution(self):
+        node = _make_node({
+            "rules": [
+                {"id": "r1", "field": "node_outputs.agent_1.output", "operator": "contains", "value": "hello"},
+            ],
+        })
+        fn = assertion_factory(node)
+        result = fn({"node_outputs": {"agent_1": {"output": "hello world"}}})
+
+        assert result["_route"] == "pass"
+        assert result["output"]["passed"] is True
+
+    def test_check_description_format(self):
+        node = _make_node({
+            "rules": [
+                {"id": "r1", "field": "status", "operator": "equals", "value": "ok"},
+            ],
+        })
+        fn = assertion_factory(node)
+        result = fn({"status": "ok"})
+
+        check = result["output"]["results"][0]
+        assert check["check"] == "status equals ok"
+        assert check["expected"] == "ok"
+        assert check["actual"] == "ok"