From 6f7121372dac2e26af83d786e67e656b1f1f40f7 Mon Sep 17 00:00:00 2001
From: cwbcheng <cwbdawei@gmail.com>
Date: Fri, 7 Aug 2026 00:37:25 +0800
Subject: [PATCH] fix: stop queued jobs from failing in-progress babysit runs

---
 server/babysitter.test.ts | 80 +++++++++++++++++++++++++++++++++++++++
 server/babysitter.ts      | 54 ++++++++++++++++----------
 2 files changed, 114 insertions(+), 20 deletions(-)

diff --git a/server/babysitter.test.ts b/server/babysitter.test.ts
index 7edd886..44e223e 100644
--- a/server/babysitter.test.ts
+++ b/server/babysitter.test.ts
@@ -6025,6 +6025,86 @@ test("babysitPR logs metadata when skipping a duplicate in-progress run", async
   assert.equal(typeof skipLog?.metadata?.activeAgeMs, "number");
 });
 
+test("runQueuedBabysitPR does not mark a live in-progress run failed when a queued job fires before replay context is persisted", async () => {
+  const storage = new MemStorage();
+  await storage.updateConfig({ autoUpdateDocs: false });
+  const pr = await storage.addPR({
+    number: 107,
+    title: "Live run PR",
+    repo: "alex-morgan-o/lolodex",
+    branch: "feature/live",
+    author: "octocat",
+    url: "https://github.com/alex-morgan-o/lolodex/pull/107",
+    status: "watching",
+    feedbackItems: [],
+    accepted: 0,
+    rejected: 0,
+    flagged: 0,
+    testsPassed: null,
+    lintPassed: null,
+    lastChecked: null,
+  });
+  let releaseFetch!: () => void;
+  let markFetchStarted!: () => void;
+  const fetchStarted = new Promise<void>((resolve) => {
+    markFetchStarted = resolve;
+  });
+  const fetchReleased = new Promise<void>((resolve) => {
+    releaseFetch = resolve;
+  });
+
+  const babysitter = new PRBabysitter(
+    storage,
+    makeWatcherGitHubService({
+      fetchPullSummary: async () => {
+        markFetchStarted();
+        await fetchReleased;
+        return makePullSummary(pr);
+      },
+      listFailingStatuses: async () => [],
+    }),
+    {
+      resolveAgent: async () => "codex",
+      ciPollIntervalMs: 0,
+      evaluateFixNecessityWithAgent: async () => ({
+        needsFix: false,
+        reason: "No code change needed",
+      }),
+      applyFixesWithAgent: async () => ({ code: 0, stdout: "", stderr: "" }),
+      runCommand: makeGitRunCommand(),
+    },
+  );
+
+  const firstRun = babysitter.babysitPR(pr.id, "codex");
+  await fetchStarted;
+
+  const runsBefore = await storage.listAgentRuns({ prId: pr.id });
+  assert.equal(runsBefore.length, 1);
+  assert.equal(runsBefore[0]?.status, "running");
+  assert.equal(runsBefore[0]?.prompt, null);
+
+  // A watcher-triggered babysit_pr job firing while the run is still live must
+  // not mark the run failed just because its replay context is not persisted yet.
+  await babysitter.runQueuedBabysitPR(pr.id, "codex");
+
+  const runsAfter = await storage.listAgentRuns({ prId: pr.id });
+  assert.equal(runsAfter.length, 1);
+  assert.equal(runsAfter[0]?.status, "running");
+  assert.equal(runsAfter[0]?.lastError, null);
+
+  const logs = await storage.getLogs(pr.id);
+  assert.ok(logs.some((log) =>
+    log.phase === "run"
+    && log.message.includes("another run is already in progress")
+  ));
+
+  releaseFetch();
+  await firstRun;
+
+  const run = await storage.getAgentRun(runsBefore[0]!.id);
+  assert.equal(run?.status, "completed");
+});
+
 test("runQueuedBabysitPR archives merged PRs instead of running or rescheduling work", async () => {
   const storage = new MemStorage();
   await storage.updateConfig({ autoUpdateDocs: false });
diff --git a/server/babysitter.ts b/server/babysitter.ts
index f0cd197..0f05406 100644
--- a/server/babysitter.ts
+++ b/server/babysitter.ts
@@ -2167,12 +2167,45 @@ export class PRBabysitter {
     }
   }
 
+  private async isRunAlreadyInProgress(
+    prId: string,
+    requestedPreferredAgent: CodingAgent,
+  ): Promise<boolean> {
+    const activeRun = this.inProgress.get(prId);
+    if (!activeRun) {
+      return false;
+    }
+    const pr = await this.storage.getPR(prId);
+    if (pr) {
+      const activeStartedAtMs = Date.parse(activeRun.startedAt);
+      const activeAgeMs = Number.isNaN(activeStartedAtMs)
+        ? null
+        : Math.max(0, this.now().getTime() - activeStartedAtMs);
+      await this.storage.addLog(pr.id, "warn", "PR work skipped because another run is already in progress", {
+        phase: "run",
+        metadata: {
+          reason: "in_progress",
+          activeRunId: activeRun.runId,
+          activePreferredAgent: activeRun.preferredAgent,
+          activeStartedAt: activeRun.startedAt,
+          activeAgeMs,
+          requestedPreferredAgent,
+        },
+      });
+    }
+    return true;
+  }
+
   async runQueuedBabysitPR(
     prId: string,
     preferredAgent: CodingAgent,
     agentSettings?: AgentRuntimeSettings,
     jobAttemptCount?: number,
   ): Promise<void> {
+    if (await this.isRunAlreadyInProgress(prId, preferredAgent)) {
+      return;
+    }
+
     const interruptedRun = (await this.storage.listAgentRuns({ status: "running", prId }))
       .slice()
       .sort((a, b) => Date.parse(b.updatedAt) - Date.parse(a.updatedAt))[0];
@@ -3137,26 +3170,7 @@ export class PRBabysitter {
       return;
     }
 
-    const activeRun = this.inProgress.get(prId);
-    if (activeRun) {
-      const pr = await this.storage.getPR(prId);
-      if (pr) {
-        const activeStartedAtMs = Date.parse(activeRun.startedAt);
-        const activeAgeMs = Number.isNaN(activeStartedAtMs)
-          ? null
-          : Math.max(0, this.now().getTime() - activeStartedAtMs);
-        await this.storage.addLog(pr.id, "warn", "PR work skipped because another run is already in progress", {
-          phase: "run",
-          metadata: {
-            reason: "in_progress",
-            activeRunId: activeRun.runId,
-            activePreferredAgent: activeRun.preferredAgent,
-            activeStartedAt: activeRun.startedAt,
-            activeAgeMs,
-            requestedPreferredAgent: preferredAgent,
-          },
-        });
-      }
+    if (await this.isRunAlreadyInProgress(prId, preferredAgent)) {
       return;
     }