openclaw - 💡(How to fix) Fix Feature: expose before_prompt_build modifying-hook timeout as config [1 comments, 2 participants]

Summary

Expose the outer plugin hook-runner timeout for modifying hooks (especially before_prompt_build) as configuration instead of keeping it hardcoded at 15s.

Problem

active-memory has its own timeoutMs config, but in current installed builds there is also an outer hook-runner deadline for modifying hooks. For before_prompt_build, that deadline is hardcoded to ~15s. In practice this means a plugin can configure a longer model/recall timeout, but the hook still gets aborted by the outer hook cap first.

This is especially visible with Active Memory recall models that sometimes complete in ~10-12s but occasionally need slightly longer. The plugin-level config gives the impression that the budget is tunable, but the actual foreground hook budget is capped elsewhere.

Example local behavior:

active-memory ... activeModel=qwen3.6-active-memory:latest done status=ok elapsedMs=9967 summaryChars=114
active-memory ... activeModel=qwen3.6-active-memory:latest done status=ok elapsedMs=10449 summaryChars=115
active-memory ... activeModel=qwen3.6-active-memory:latest done status=timeout elapsedMs=15189 summaryChars=0
active-memory ... activeModel=qwen3.6-active-memory:latest done status=timeout elapsedMs=15492 summaryChars=0

The plugin config was set to timeoutMs: 15000; raising it would not solve the real limit if the hook runner still enforces a hard 15s before_prompt_build modifying-hook deadline.

Proposed solution

Add a documented config path for the modifying-hook runner timeout, ideally with per-hook and/or per-plugin override support.

Possible shape:

{
  "plugins": {
    "hookTimeouts": {
      "before_prompt_build": 30000
    },
    "entries": {
      "active-memory": {
        "hooks": {
          "before_prompt_build": {
            "timeoutMs": 30000,
            "onTimeout": "skip" // fail open for recall/context plugins
          }
        }
      }
    }
  }
}

Exact schema is not important; the important part is that operators can tune the outer hook cap without patching OpenClaw's bundled JS.

Desired behavior

• before_prompt_build modifying-hook timeout is configurable.
• Defaults remain safe and responsive for normal users.
• Context/recall plugins such as Active Memory should fail open on timeout unless explicitly configured otherwise.
• Logs should identify whether timeout came from plugin-level timeoutMs or the outer hook-runner cap.
• Documentation should clarify the relationship between Active Memory timeoutMs and hook-runner timeout.

Why this matters

Active Memory is useful in direct chat, but model latency varies. A hardcoded outer 15s cap forces users to choose only models that reliably finish under 15s, even if they are willing to trade a slightly longer foreground budget for better recall quality. Making the cap configurable avoids local code patches and makes the behavior explicit.

Related issues

• #62702 - broader non-blocking async plugin hooks / configurable timeout work
• #65309 - Active Memory blocks direct-chat turns and times out across model configs
• #69182 - Active Memory can block chat replies
• #66804 / #73306 - Active Memory 15s timeout reports