openclaw - 💡(How to fix) Fix Recurring Anthropic streaming JSON parse errors (Sonnet 4.5 / Opus) — failover often fails to recover [1 comments, 2 participants]

embedded_run_failover_decision: stage=assistant, decision=fallback_model, model=claude-sonnet-4-5, rawErrorPreview="Expected ',' or '}' after property value in JSON at position 160"

lane task error: lane=session:agent:main:main durationMs=60457 error="FailoverError: Expected ',' or '}' after property value in JSON at position 160"

embedded_run_agent_end: model=claude-sonnet-4-5, isError=true, failoverReason=null, error="Expected ',' or '}' after property value in JSON at position 162"

TL;DR

Implement a retry mechanism for JSON parse errors to trigger failover consistently, ensuring raw errors are not surfaced to users.

Guidance

• Review the failoverReason logic to handle null cases, allowing failover to occur for all JSON parse errors, not just timeouts.
• Modify the error handling to catch and retry JSON parse errors, preventing them from being exposed to users.
• Consider enhancing the streaming JSON parser to be more resilient to truncated or malformed chunks, potentially by implementing a buffer or more robust parsing mechanism.
• Investigate the related issues (#59080, #46370, #33280, #60002) to identify potential fixes or workarounds for similar JSON parsing problems.

Example

No specific code snippet can be provided without more context, but a possible approach might involve wrapping the JSON parsing code in a try-catch block and retrying the operation on failure:

try {
  // JSON parsing code
} catch (error) {
  if (error instanceof JSONParseError) {
    // Trigger retry or failover
  } else {
    // Handle other error types
  }
}

Notes

The provided information suggests that the issue is related to the JSON parsing mechanism and the failover logic. However, without more details about the implementation, it's difficult to provide a more specific solution.

Recommendation

Apply a workaround by implementing a retry mechanism for JSON parse errors to ensure consistent failover and prevent raw errors from being surfaced to users. This approach addresses the immediate issue and provides a more robust error handling mechanism.