openclaw - 💡(How to fix) Fix feat: Add silentErrorReplies / suppressAgentErrors support for WhatsApp channel [1 participants]

When the primary model (e.g., Anthropic Claude) is rate-limited or overloaded and the fallback model also fails, OpenClaw sends a generic error message to the user:

• buildExternalRunFailureText() in agent-runner.runtime always returns the error message

1. messages.silentErrorReplies: true — suppress error messages entirely (silent fail)
2. messages.errorMessage: "..." — custom error text instead of the generic one

• WhatsApp DMs with business contacts who receive confusing error messages

TL;DR

Implement a config option to suppress or customize the agent failure message on WhatsApp, such as messages.silentErrorReplies or messages.errorMessage, to improve user experience.

Guidance

• Introduce a new config option under messages to control error message behavior on WhatsApp, e.g., messages.whatsapp.silentErrorReplies or messages.whatsapp.errorMessage.
• Modify buildExternalRunFailureText() in agent-runner.runtime to respect the new config option and suppress or customize the error message accordingly.
• Consider adding per-channel support for silentErrorReplies to allow for more flexibility in error handling across different channels.
• Update the logging mechanism to surface errors to the operator (logs) instead of the end user, ensuring silent retries.

Example

// Example config snippet
messages: {
  whatsapp: {
    silentErrorReplies: true, // or
    errorMessage: "Sorry, we're experiencing technical difficulties. Please try again later."
  }
}

Notes

The proposed solution assumes that the existing silentErrorReplies feature in the Telegram config path can be extended to WhatsApp, and that the buildExternalRunFailureText() function can be modified to respect the new config option.

Recommendation

Apply a workaround by introducing a custom config option to suppress or customize the error message on WhatsApp, as this will allow for a more user-friendly experience while the primary and fallback models are rate-limited or overloaded.