openclaw - 💡(How to fix) Fix Anthropic Max usage alert text blocks delivered as assistant messages to channels [1 comments, 2 participants]

TL;DR

Strip text content blocks matching the Anthropic Max usage alert pattern from the response pipeline before delivery to prevent unwanted billing metadata from being stored and sent as bot messages.

Guidance

• Identify the response pipeline where the usage alert blocks are being injected and modify it to filter out messages matching the pattern /⚠️ \*\*Claude Max Usage Alert\*\*[\s\S]*?Extra: \$[\d.]+\/\$[\d.]+/.
• Consider adding a configurable messages.outputStripPatterns array to allow for easy modification of the filter patterns in the future.
• Verify that the filtering is working correctly by checking the session JSONL and channel outputs for the absence of usage alert blocks.
• Test the fix with different types of responses and edge cases to ensure it does not inadvertently filter out legitimate messages.

Example

const filterPattern = /⚠️ \*\*Claude Max Usage Alert\*\*[\s\S]*?Extra: \$[\d.]+\/\$[\d.]+/;
const filteredMessages = messages.filter(message => !filterPattern.test(message.content));

Notes

This fix assumes that the usage alert blocks always follow the specified pattern and that filtering them out will not affect the functionality of the bot. If the pattern changes or if there are other types of messages that need to be filtered, the messages.outputStripPatterns array can be modified accordingly.

Recommendation

Apply workaround by filtering out the usage alert blocks using the specified pattern, as this is a targeted solution that addresses the specific issue without requiring an upgrade or modification of the underlying system.