UserPromptSubmit hook exits with code 1 when worker is unreachable, causing "Failed with non-blocking status code" error on every prompt

[nitin-vavdiya]

Before submitting

• I searched existing issues and confirmed this is not a duplicate

---

📝 Manual Bug Report

Bug Description

The UserPromptSubmit hook command ends with:

[ "$_HEALTH" = "1" ] && node "$_R/scripts/bun-runner.js" "$_R/scripts/worker-service.cjs" hook claude-code session-init

When _HEALTH=0 (worker unreachable or health check times out), the [ test evaluates false and exits with code 1. Claude Code treats this as a hook failure and shows the following error on every user prompt:

UserPromptSubmit hook error
Failed with non-blocking status code: No stderr output

No stderr output is produced because the shell [ command exits silently with code 1 — leaving the user with no indication of what went wrong.

---

Steps to Reproduce

1. Worker is not running (first session, after crash, or port mismatch after upgrade)
2. Claude Code fires UserPromptSubmit hook
3. Health check loop retries 10 times and fails → _HEALTH=0
4. [ "0" = "1" ] && node ... → exits code 1
5. Claude Code reports hook error on every subsequent prompt

Port mismatch scenario (also triggers this)

• ~/.claude-mem/settings.json has CLAUDE_MEM_WORKER_PORT: "37777" (set via UI)
• After upgrading to a new version, the old daemon (e.g. 12.3.8) keeps running on 37777
• New version's hook falls back to UID-based port: 37700 + uid % 100
• On this system: 37700 + 502 % 100 = 37702 — mismatch → health check fails → exit code 1

---

Expected Behavior

Hook exits 0 when worker is unreachable. The health check already makes this a graceful no-op — failing to reach the worker should not be reported as a hook error.

Minimal fix:

# Before
[ "$_HEALTH" = "1" ] && node ... hook claude-code session-init

# After — option A
[ "$_HEALTH" = "1" ] && node ... hook claude-code session-init || true

# After — option B
if [ "$_HEALTH" = "1" ]; then
  node ... hook claude-code session-init
fi

---

Environment

Field	Value
claude-mem version	12.3.9
Claude Code version	2.1.118
OS	macOS 26.2 (Darwin 25.2.0)
Platform	Apple Silicon (arm64)
Shell	zsh
User UID	502 → fallback port 37702, settings port 37777

---

Logs

Worker itself is healthy once running — the error is purely in hook exit-code handling:

[2026-04-23 11:38:33.067] [INFO ] [SYSTEM] Worker became healthy while waiting on live PID
[2026-04-23 11:38:33.196] [INFO ] [HTTP  ] SessionRoutes: handleSessionInitByClaudeId called
[2026-04-23 11:38:33.198] [INFO ] [HOOK  ] INIT_COMPLETE | skipped_agent_init=true | reason=context_already_injected

---

Additional Context

Two related but separable issues:

1. Exit code bug — [ "$_HEALTH" = "1" ] && ... exits 1 when worker is down. Simple || true fix. Affects anyone whose worker is slow to start or temporarily unavailable.

2. Port mismatch after upgrade — Old daemon from a previous version stays alive on the port it was started with. New version's hook calculates a different fallback port. Neither side is "wrong" but they disagree, triggering issue #1.

Workaround applied: Set CLAUDE_MEM_WORKER_PORT in ~/.claude-mem/settings.json to match the UID-based fallback (37700 + uid % 100), then restart the worker.

[github-actions]

The GitHub issue highlights a problem with the UserPromptSubmit hook in the claude-mem project, where the hook exits with code 1 when the worker is unreachable, leading to an error being displayed for every user prompt. This is caused by a faulty condition in the shell script ([ "$_HEALTH" = "1" ] && ...) that evaluates false and exits silently, offering no stderr output or explanation for the failure. The issue can occur due to slow worker initialization, health-check timeouts, or port mismatches after upgrades, where the new version defaults to a UID-based port but an old running daemon uses a previous port setting. The proposed fix is to modify the shell script to ensure the hook exits with code 0 when the worker is unavailable, either by appending || true or refactoring the condition into an if statement. The reporter also highlights a workaround involving manually adjusting the port in the settings file to align with

[matheusteixeira7]

Same class of bug triggers on SessionStart:startup (claude-mem 12.3.9), not just UserPromptSubmit. The error shows up as:

SessionStart:startup hook error
Failed with non-blocking status code: No stderr output

In my case port alignment is fine (settings.json = worker.pid = 37777, /health returns 200, no errors in ~/.claude-mem/logs/), so the port-mismatch scenario doesn't explain it — looks like a startup race where one of the three SessionStart:startup commands returns exit ≠ 0 before the worker is ready.

Suggest applying the same || true / explicit exit 0 guard from the proposed fix to all three SessionStart:startup entries in hooks/hooks.json, not only UserPromptSubmit.