SessionStart hook fails silently with 'No stderr output' when worker health check times out

[cags84]

Summary

The SessionStart hook in hooks/hooks.json (worker health check block) exits with code 1 without writing to stderr when the bun worker on localhost:37777 doesn't respond to /health within the 8-second retry window. Claude Code surfaces this as a confusing non-blocking error:

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

Repro

1. Ensure no claude-mem worker is running (lsof -iTCP:37777 -sTCP:LISTEN returns empty).
2. Start a new Claude Code session.
3. If the worker takes more than ~8s to become healthy, the hook fails silently.

Also reproducible after system sleep/reboot when the previous worker process died and the new one is slow to bind the port.

Root Cause

In hooks/hooks.json, the second and third SessionStart hooks end with:

for i in 1 2 3 4 5 6 7 8; do curl -sf http://localhost:37777/health >/dev/null 2>&1 && break; sleep 1; done; curl -sf http://localhost:37777/health >/dev/null 2>&1 || exit 1

The exit 1 bails silently — no diagnostic message, no log path, nothing. The user sees only the generic Claude Code message with no way to know which plugin or hook failed.

Suggested Fixes

1. Emit a clear stderr message before exiting, e.g.:

   || { echo "claude-mem: worker on :37777 did not become healthy within 8s" >&2; exit 1; }

2. Consider increasing the retry budget, or using exponential backoff, for cold-start scenarios.
3. Optionally log the worker startup output to a file (~/.claude-mem/worker.log) so users can debug without reproducing.

Environment

• macOS Darwin 25.4.0
• Claude Code (CLI)
• claude-mem plugin version 12.1.0
• Worker eventually becomes healthy; this is purely a startup race.

Workaround

Ignore the message (it's non-blocking) — subsequent sessions work fine once the worker is up.

---

Reported by @cags84

[github-actions]

A GitHub issue reports that the SessionStart hook for the Claude Code CLI fails silently with no stderr output when the worker health check on localhost:37777 times out during its 8-second retry window. This occurs if the worker is slow to bind the port after a system reboot, sleep, or when no active worker is running. Although the issue is non-blocking, users only encounter a generic error message "No stderr output," leaving them unable to trace which hook or plugin caused the failure. The root cause lies in the hook's shell script (hooks/hooks.json), which exits with code 1 without emitting any diagnostic message. The reporter suggests improvements, such as logging clear errors to stderr, increasing the retry duration using exponential backoff, or capturing worker startup logs for debugging. A workaround exists—ignoring the error and waiting for the worker to become healthy—but the issue highlights a need for better clarity in error messaging during startup.

[thedotmack]

Closed during the April 2026 backlog cleanup. This was verified already-fixed in v12.1.1 HEAD. Please update to the latest build. If you still see the issue on current version, open a fresh ticket with repro.