feat: add dry-run workflow execution boundary

2026-07-24 22:30:41 +08:00
parent 646178944d
commit c9ad841543
12 changed files with 553 additions and 17 deletions
@@ -468,6 +468,63 @@ test('shadow observer failure cannot fail a Native run and chat runs are not obs
  assert.equal(JSON.parse(failure.dataJson).code, 'SHADOW_UNAVAILABLE');
 });

+test('canary dry-run records the candidate decision without changing Native run state', async () => {
+  const pool = createFakePool();
+  const gateway = createAgentRunGateway({
+    pool,
+    userAuth: {},
+    tkmindProxy: {},
+    autoDispatch: false,
+    observeWorkflowRun: async () => ({
+      observed: false,
+      mode: 'canary',
+      executionPlan: {
+        version: 'workflow-execution-decision-v1',
+        candidateEngine: 'langgraph',
+        effectiveEngine: 'native',
+        fallbackEngine: 'native',
+        reason: 'execution_gate_disabled',
+        candidateReason: 'percentage',
+        configVersion: 9,
+        bucket: 3,
+        dryRun: true,
+        handoffAllowed: false,
+      },
+    }),
+  });
+
+  const run = await gateway.createRun('user-canary', {
+    requestId: 'req-canary-dry-run',
+    sessionId: 'session-canary',
+    userMessage: { role: 'user', content: 'refactor this module' },
+    toolMode: 'code',
+    taskType: 'repo_refactor',
+  });
+
+  await waitFor(() => pool.events.some(
+    (event) => event.eventType === 'workflow_execution_planned',
+  ));
+  assert.equal(run.status, 'queued');
+  assert.equal(
+    pool.events.some((event) => event.eventType === 'workflow_shadow_completed'),
+    false,
+  );
+  const event = pool.events.find((item) => item.eventType === 'workflow_execution_planned');
+  assert.deepEqual(JSON.parse(event.dataJson), {
+    version: 'workflow-execution-decision-v1',
+    mode: 'canary',
+    candidateEngine: 'langgraph',
+    effectiveEngine: 'native',
+    fallbackEngine: 'native',
+    reason: 'execution_gate_disabled',
+    candidateReason: 'percentage',
+    configVersion: 9,
+    bucket: 3,
+    dryRun: true,
+    handoffAllowed: false,
+  });
+});
+
 test('agent run starts a session and marks submitted reply as succeeded', async () => {
  const pool = createFakePool();
  const submitted = [];