Stop hook running in cycle after response (v2.3)

[orestesgarcia]

First, I want to say thank you Daniel for this amazing contribution - I am a fan of your work!

Issue

After installing PAI released version 2.3, every time the agent runs and gives me a response, it goes into a cycle where the stop hook seems to be running repeatedly and printing weird lines to stdio.

Screenshot

In the attached screenshot, you can see this specifically ran for 38 seconds where the stop hook keeps cycling:

The output shows repeated lines like:

* Zesting... (ctrl+c to interrupt · running stop hook · 37s · ↓ 448 tokens · thought for 2s)
n mode

This repeats multiple times before finally completing.

Environment

• PAI Version: 2.3
• Installation: Fresh install following documentation

Expected Behavior

The stop hook should run once and complete quickly after a response.

Actual Behavior

The stop hook appears to run in a cycle for extended periods (38+ seconds), printing repeated "Zesting..." messages with "n mode" fragments.

Questions

Any idea why this might be happening? Is there a configuration issue or known workaround?

---

[genebushmaster]

same same on v2.4. Do you have all dependencies API keys configured? If not, there are no timeouts in the hooks code for elevenlabs and ntfy and they might be causing the hang.
In my case, root cause was observability server running without a client:

1. StopOrchestrator calls handlers via Promise.allSettled
2. One handler (ResponseCapture) calls sendEventToObservability()
3. That fetch tries to connect to localhost:4000
4. Nothing listening → TCP connection stuck in SYN_SENT
5. No timeout → hangs indefinitely
6. Promise.allSettled waits for ALL promises → entire hook blocked

[docxology]

Fix: AbortController timeouts on all network fetches in the Stop hook chain

@genebushmaster's root cause analysis is correct. The sendEventToObservability() fetch to localhost:4000 has no timeout, and when nothing is listening, the TCP connection can hang indefinitely — blocking the entire Promise.allSettled in StopOrchestrator.

We audited every network call in the Stop hook chain and found 5 fetch calls with no timeout across 4 files. Here's the fix for each:

1. hooks/lib/observability.ts — Primary culprit (2s timeout)

 export async function sendEventToObservability(event: ObservabilityEvent): Promise<void> {
   try {
+    const controller = new AbortController();
+    const timeout = setTimeout(() => controller.abort(), 2000);
     const response = await fetch('http://localhost:4000/events', {
       method: 'POST',
       headers: {
         'Content-Type': 'application/json',
         'User-Agent': 'PAI-Hook/1.0'
       },
       body: JSON.stringify(event),
+      signal: controller.signal,
     });
+    clearTimeout(timeout);

2. hooks/handlers/VoiceNotification.ts — Voice server fetch (5s timeout)

This one is awaited inside Promise.allSettled in the orchestrator, so if the voice server at localhost:8888 isn't running, it blocks too.

+    const controller = new AbortController();
+    const timeout = setTimeout(() => controller.abort(), 5000);
     const response = await fetch('http://localhost:8888/notify', {
       method: 'POST',
       headers: { 'Content-Type': 'application/json' },
       body: JSON.stringify(payload),
+      signal: controller.signal,
     });
+    clearTimeout(timeout);

3. hooks/handlers/SystemIntegrity.ts — Voice notification for integrity check (5s timeout)

Same localhost:8888 fetch, fire-and-forget but leaks resources without a timeout.

4. hooks/lib/notifications.ts — ntfy.sh + Discord webhook fetches (5s timeout each)

External network calls that could hang under poor connectivity.

Result

Worst-case StopOrchestrator completion time goes from indefinite (30-38s+ observed) to ~5 seconds max.

Note: Separate intermittent non-firing issue

There's also a Claude Code platform bug where the Stop hook doesn't fire consistently on macOS — tracked in anthropics/claude-code#22925. That's a separate issue from the hanging/cycling reported here and can't be fixed in PAI code. The timeout fix above addresses the cycling when the hook does fire.