From f9f798f32070e753ef11abba6b6609419af614f5 Mon Sep 17 00:00:00 2001
From: Dotta <bippadotta@protonmail.com>
Date: Thu, 3 Sep 2026 08:34:08 -0500
Subject: [PATCH] test(runner): allow accepted plan ACPX session resets

---
 tests/runner-e2e/run-observations.ts | 55 +++++++++++++++++++---------
 tests/runner-e2e/runner.spec.ts      | 11 ++++++
 tests/runner-e2e/support.test.ts     | 19 ++++++++++
 3 files changed, 68 insertions(+), 17 deletions(-)

diff --git a/tests/runner-e2e/run-observations.ts b/tests/runner-e2e/run-observations.ts
index f8883ad6b0..90d4312f19 100644
--- a/tests/runner-e2e/run-observations.ts
+++ b/tests/runner-e2e/run-observations.ts
@@ -15,6 +15,37 @@ export interface ObservableProviderSessionRun {
   contextSnapshot?: Record<string, unknown> | null;
 }
 
+export function acceptedPlanSessionResetFailures(
+  provider: "codex" | "opencode" | "acpx",
+  previousSessionId: string | null | undefined,
+  current: ObservableProviderSessionRun,
+): string[] | null {
+  const context = record(current.contextSnapshot);
+  const acceptedPlanReset =
+    context.forceFreshSession === true &&
+    context.workspaceRefreshReason === "accepted_plan_confirmation" &&
+    context.source === "issue.interaction.accept" &&
+    context.interactionStatus === "accepted";
+  if (!acceptedPlanReset) return null;
+
+  const failures: string[] = [];
+  if (current.sessionIdBefore) {
+    failures.push(
+      `expected accepted Plan run ${current.id} to start without a prior provider session`,
+    );
+  }
+  if (
+    previousSessionId &&
+    current.sessionIdAfter &&
+    current.sessionIdAfter === previousSessionId
+  ) {
+    failures.push(
+      `expected accepted Plan run ${current.id} to rotate the ${provider} provider session`,
+    );
+  }
+  return failures;
+}
+
 function record(value: unknown): Record<string, unknown> {
   return value && typeof value === "object" && !Array.isArray(value)
     ? (value as Record<string, unknown>)
@@ -89,23 +120,13 @@ export function providerSessionContinuityFailures(
     if (index === 0) continue;
 
     const previousSessionId = runs[index - 1]?.sessionIdAfter;
-    const context = record(current.contextSnapshot);
-    const acceptedPlanReset =
-      context.forceFreshSession === true &&
-      context.workspaceRefreshReason === "accepted_plan_confirmation" &&
-      context.source === "issue.interaction.accept" &&
-      context.interactionStatus === "accepted";
-    if (acceptedPlanReset) {
-      if (current.sessionIdBefore) {
-        failures.push(
-          `expected accepted Plan run ${current.id} to start without a prior provider session`,
-        );
-      }
-      if (previousSessionId && currentSessionId === previousSessionId) {
-        failures.push(
-          `expected accepted Plan run ${current.id} to rotate the ${provider} provider session`,
-        );
-      }
+    const acceptedPlanResetFailures = acceptedPlanSessionResetFailures(
+      provider,
+      previousSessionId,
+      current,
+    );
+    if (acceptedPlanResetFailures) {
+      failures.push(...acceptedPlanResetFailures);
       continue;
     }
 
diff --git a/tests/runner-e2e/runner.spec.ts b/tests/runner-e2e/runner.spec.ts
index 2ac70bd877..a21b45a50e 100644
--- a/tests/runner-e2e/runner.spec.ts
+++ b/tests/runner-e2e/runner.spec.ts
@@ -10,6 +10,7 @@ import { runnerE2EServerControlPaths } from "./harness-env.js";
 import { setupLiveFixtures, type LiveFixtureValues } from "./live-fixtures.js";
 import { evaluateMatcher, type MatcherResult } from "./matchers.js";
 import {
+  acceptedPlanSessionResetFailures,
   isControlPlaneGovernedResponseWait,
   isNonExecutingReviewFenceRun,
   numberedPlanStepCount,
@@ -1542,6 +1543,16 @@ for (const execution of executions) {
               );
               continue;
             }
+            const acceptedPlanResetFailures =
+              acceptedPlanSessionResetFailures(
+                "acpx",
+                previousSessionId,
+                current,
+              );
+            if (acceptedPlanResetFailures) {
+              invariantFailures.push(...acceptedPlanResetFailures);
+              continue;
+            }
             if (previousSessionId === currentSessionId) continue;
             const currentEvents =
               runEventsByRun.find((captured) => captured.runId === current.id)
diff --git a/tests/runner-e2e/support.test.ts b/tests/runner-e2e/support.test.ts
index d966aac661..ab7983e517 100644
--- a/tests/runner-e2e/support.test.ts
+++ b/tests/runner-e2e/support.test.ts
@@ -8,6 +8,7 @@ import { RunnerApi } from "./api.js";
 import { FixtureRegistry } from "./fixture-registry.js";
 import { classifyFailure, shouldRetryFailure } from "./failure-classifier.js";
 import {
+  acceptedPlanSessionResetFailures,
   assertIsolatedServerEnvironment,
   buildPaperclipServerEnvironment,
   resolvePaperclipRunnerBinaryForHarness,
@@ -415,6 +416,24 @@ describe("runner E2E run observations", () => {
     ).toEqual([
       "expected codex to preserve its provider session for run resumed",
     ]);
+    expect(
+      acceptedPlanSessionResetFailures(
+        "acpx",
+        initial.sessionIdAfter,
+        acceptedPlan,
+      ),
+    ).toEqual([]);
+    expect(
+      acceptedPlanSessionResetFailures("acpx", initial.sessionIdAfter, {
+        ...acceptedPlan,
+        sessionIdBefore: initial.sessionIdAfter,
+      }),
+    ).toEqual([
+      "expected accepted Plan run accepted-plan to start without a prior provider session",
+    ]);
+    expect(
+      acceptedPlanSessionResetFailures("acpx", initial.sessionIdAfter, resumed),
+    ).toBeNull();
   });
 });