openclaw - 💡(How to fix) Fix Message duplication: agent runner retries full turns on tool errors, resending text to user

When a tool call fails with an error (e.g., "guildId required" on Discord message send, or "node not connected"), the agent runner retries the entire turn. Since the model generates both text output AND a tool call in the same turn, the text gets delivered to the user again on each retry. This results in 7-10+ identical messages per user input.

• The file has transient error detection for HTTP errors (429, 5xx, quota) via TRANSIENT_ERROR_DETAIL_HINT_RE
• The agent should either: (a) abort the turn and report the error, or (b) skip the failed tool and deliver the text response once

TL;DR

Implement deduplication logic or a max-turn limit for repeated tool failures to prevent duplicate messages from being sent to the user.

Guidance

• Identify and handle terminal tool errors (e.g., "guildId required", "node not connected") separately from transient errors to prevent retries.
• Consider implementing a retry count limit or exponential backoff for tool failures to prevent infinite loops.
• Modify the agent runner to skip the failed tool and deliver the text response only once, instead of regenerating the full response on each retry.
• Review the TRANSIENT_ERROR_DETAIL_HINT_RE pattern in agent-runner.runtime-CTlghBhJ.js to ensure it correctly handles terminal tool errors.

Example

No code snippet is provided as the issue does not contain sufficient code context.

Notes

The provided information suggests that the issue is specific to the agent runner and tool error handling. The solution may require modifications to the agent-runner.runtime-CTlghBhJ.js file or the underlying logic for handling tool failures.

Recommendation

Apply a workaround by modifying the agent runner to handle terminal tool errors and implement a retry count limit or exponential backoff, as upgrading to a fixed version is not mentioned in the issue.