Hook infinite retry loop on summarize event (Permission denied)

[HN84956371-code]

Bug Description

When the claude-mem worker becomes unreachable during a session, the summarize hook enters an infinite retry loop that never stops. Each retry generates a new "Stop hook feedback" message that triggers another Claude response, which triggers another hook attempt, creating an endless cycle.

Environment

• OS: Windows 11 Home
• Claude Code: v2.1.169
• claude-mem: v13.4.0
• Shell: Bash (via Git for Windows)

Steps to Reproduce

1. Start a Claude Code session with claude-mem installed
2. The worker becomes unreachable (in this case due to printf: write error: Permission denied)
3. Any tool call triggers the hook, which fails
4. On session end/summarize, the hook retries indefinitely

Observed Behavior

The hook script fails with:

/usr/bin/bash: line 1: printf: write error: Permission denied
claude-mem worker unreachable for N consecutive hooks.

Where N increments from 1 to 33+ without stopping. The summarize event hook keeps retrying on every Claude response, creating an infinite feedback loop that makes the session unusable.

Expected Behavior

After N consecutive failures (e.g., 3-5), the hook should stop retrying for the remainder of the session, or at minimum not retry on the summarize event which creates a feedback loop.

Impact

• Wastes API tokens (each retry triggers a Claude response)
• Makes the session difficult to close gracefully
• User must force-quit with Ctrl+C

Suggestion

Add a max retry cap or exponential backoff. If the worker is unreachable after N attempts, disable hooks for the rest of the session and log a single warning instead of retrying indefinitely.

[chukui2005]

I am also affected by this issue, here is my environment.

• OS: Windows 11 Pro
• Claude Code: v2.1.132
• claude-mem: v13.4.2
• Shell: Bash (via Git for Windows)

[thedotmack]

Update: #2865 landed the root-cause fix for the printf: write error: Permission denied trigger (the candidate-pipe break in the hook shell prelude closed the pipe early; Cygwin/MSYS reports the producer's next write as EACCES). That removes the proximate cause of the failing summarize hook. The second half of this report — the hook retrying in an infinite loop instead of capping/backing off — is a separate defect in the retry logic and remains open; leaving this issue open to track it.

[thedotmack]

Consolidating into #3605 (plan-17). The root cause and fix sequencing are tracked there alongside the rest of the cluster — please follow that issue for progress.