Stop hook loops infinitely due to {"continue":true} response

[MistySun19]

Bug Description

The stop hook (worker-service.cjs hook claude-code summarize) causes an infinite loop at session end. Claude Code keeps re-triggering the stop hook, requiring manual interruption (Ctrl+C).

Environment

• Claude Code version: latest (as of 2026-03-05)
• Plugin version: 10.4.1 (manifest) / 10.4.3 (package.json)
• OS: macOS (Darwin 24.6.0)
• Node: available, Bun: available

Steps to Reproduce

1. Enable claude-mem plugin
2. Use Claude Code normally
3. When the session ends or Claude stops responding, observe the stop hooks

Observed Behavior

The stop hook fires repeatedly in an infinite loop:

⏺ Ran 3 stop hooks (ctrl+o to expand)
  ⎿ Stop hook error: [...worker-service.cjs hook claude-code summarize]: No stderr output

⏺ Ran 3 stop hooks (ctrl+o to expand)
  ⎿ Stop hook error: [...worker-service.cjs hook claude-code summarize]: No stderr output

(repeats indefinitely until interrupted)

Root Cause Analysis

Running the hook manually shows it returns:

{"continue":true,"suppressOutput":true}

The "continue":true field appears to signal Claude Code that the session should continue, which triggers another round of stop hooks, creating an infinite loop.

Expected Behavior

The stop hook should return {"continue":false} (or omit the field) so Claude Code terminates cleanly after the summarize hook completes.

Workaround

Disable the plugin in ~/.claude/settings.json:

"enabledPlugins": {
  "claude-mem@thedotmack": false
}

[github-actions]

A GitHub issue reports that the stop hook (worker-service.cjs hook claude-code summarize) in Claude Code triggers an infinite loop at the session end, requiring manual termination using Ctrl+C. The problem occurs with the claude-mem plugin enabled, where the stop hooks repeatedly fire due to the hook response {"continue":true,"suppressOutput":true} signaling Claude Code to continue the session. This behavior prevents clean termination. The expected response is {"continue":false} or omitting the "continue" field entirely to avoid re-triggering the hooks. A suggested workaround is to disable the claude-mem plugin in the settings file until the issue is resolved. This issue affects users of the latest Claude Code version (as of March 2026) with plugin version 10.4.1/10.4.3 on macOS.

[AlexWorland]

This is the same root cause as #1290. The formatOutput() in src/cli/adapters/claude-code.ts returns {"continue":true,"suppressOutput":true} which are not valid Claude Code Stop hook fields. Claude Code interprets continue: true as a signal to re-invoke the hook, causing the infinite loop.

PR #1291 fixes this by returning {} (empty object) for the default case, which correctly allows the stop to proceed without unrecognized fields.

[elenasteinvorth]

Hi @MistySun19 — I saw your issue #1288 about the infinite loop at session end. That sounds incredibly frustrating (having to Ctrl+C every time?).

I'm Elena, an entrepreneur researching the AI agent memory problem. I'm trying to understand:

1. Is claude-mem still worth using for you despite this bug?
2. Have you found any workarounds or alternatives?
3. What would make you switch to a different memory solution?
4. Is the memory benefit worth dealing with bugs like this?

I'm not selling anything — just trying to understand if the memory problem is painful enough that people tolerate significant bugs, or if there's room for a more stable alternative.

Any insights would be helpful.

Thanks,
Elena