From 42c37978798241fcc2d31f478c3e60a5a7467013 Mon Sep 17 00:00:00 2001
From: yoshitarof <yoshitarof@gmail.com>
Date: Tue, 5 May 2026 18:54:42 +0900
Subject: [PATCH] fix(stop-review-gate): treat job as inactive when its pid is
 dead

Codex plugin job state (state.json / jobs/task-*.json) records
status: "running" while a job is in flight. The dying codex.exe is
expected to update this field to "completed" / "cancelled" on exit.
But that responsibility cannot be honored if the process is killed by
something external (SIGKILL, OS reboot, power loss, a cleanup script,
or `taskkill /F`).

Result: zombie entries with status: "running" accumulate in
state.json. stop-review-gate-hook.mjs reads them, finds a "running"
entry, and emits "Codex task task-XYZ is still running." on every
session end -- even when no Codex process is actually alive.

Fix: add isPidAlive(pid) and isJobActive(job) helpers in lib/state.mjs.
isPidAlive uses process.kill(pid, 0) -- cross-platform: ESRCH = dead,
EPERM = exists but unsignalable (still alive). isJobActive combines
the existing status check with the pid liveness probe.

Replace the inline filter in stop-review-gate-hook.mjs with
jobs.find(isJobActive). The hook now only reports a running task when
it is actually running.

This is read-only -- the hook does not mutate state.json. Persistent
zombie cleanup is a separate concern that deserves its own PR.

Smoke-tested locally on Windows (Node v24): isPidAlive correctly
returns true for self pid, false for unused pid 99999999, false for
null. isJobActive correctly returns true for {running, alive pid},
false for {running, dead pid}, false for {completed, alive pid},
false for {running, no pid}.
---
 plugins/codex/scripts/lib/state.mjs           | 30 +++++++++++++++++++
 .../codex/scripts/stop-review-gate-hook.mjs   |  4 +--
 2 files changed, 32 insertions(+), 2 deletions(-)

diff --git a/plugins/codex/scripts/lib/state.mjs b/plugins/codex/scripts/lib/state.mjs
index 2da23498..11e12663 100644
--- a/plugins/codex/scripts/lib/state.mjs
+++ b/plugins/codex/scripts/lib/state.mjs
@@ -150,6 +150,36 @@ export function listJobs(cwd) {
   return loadState(cwd).jobs;
 }
 
+/**
+ * Check whether a process with the given pid is currently alive.
+ * Cross-platform: uses `process.kill(pid, 0)` which sends signal 0
+ * (no-op signal) and throws ESRCH if the process is gone. EPERM means
+ * the process exists but we lack permission to signal it (still alive).
+ */
+export function isPidAlive(pid) {
+  if (!pid || typeof pid !== "number") return false;
+  try {
+    process.kill(pid, 0);
+    return true;
+  } catch (err) {
+    return err && err.code === "EPERM";
+  }
+}
+
+/**
+ * A job counts as "active" only if its status says queued/running AND
+ * the recorded pid is still alive. This prevents zombie state from
+ * causing false alarms when the codex process was SIGKILL'd
+ * (e.g., by a cleanup script or an OS reboot) without getting a chance
+ * to write status="completed" to disk.
+ */
+export function isJobActive(job) {
+  if (!job) return false;
+  if (job.status !== "queued" && job.status !== "running") return false;
+  if (!job.pid) return false;
+  return isPidAlive(job.pid);
+}
+
 export function setConfig(cwd, key, value) {
   return updateState(cwd, (state) => {
     state.config = {
diff --git a/plugins/codex/scripts/stop-review-gate-hook.mjs b/plugins/codex/scripts/stop-review-gate-hook.mjs
index 2346bdcf..5bec0f75 100644
--- a/plugins/codex/scripts/stop-review-gate-hook.mjs
+++ b/plugins/codex/scripts/stop-review-gate-hook.mjs
@@ -8,7 +8,7 @@ import { fileURLToPath } from "node:url";
 
 import { getCodexAvailability } from "./lib/codex.mjs";
 import { loadPromptTemplate, interpolateTemplate } from "./lib/prompts.mjs";
-import { getConfig, listJobs } from "./lib/state.mjs";
+import { getConfig, isJobActive, listJobs } from "./lib/state.mjs";
 import { sortJobsNewestFirst } from "./lib/job-control.mjs";
 import { SESSION_ID_ENV } from "./lib/tracked-jobs.mjs";
 import { resolveWorkspaceRoot } from "./lib/workspace.mjs";
@@ -146,7 +146,7 @@ function main() {
   const config = getConfig(workspaceRoot);
 
   const jobs = sortJobsNewestFirst(filterJobsForCurrentSession(listJobs(workspaceRoot), input));
-  const runningJob = jobs.find((job) => job.status === "queued" || job.status === "running");
+  const runningJob = jobs.find(isJobActive);
   const runningTaskNote = runningJob
     ? `Codex task ${runningJob.id} is still running. Check /codex:status and use /codex:cancel ${runningJob.id} if you want to stop it before ending the session.`
     : null;