openclaw - 💡(How to fix) Fix Subagent runs don't use model fallback chain on 429/rate-limit errors [1 participants]

When a subagent is spawned via sessions_spawn, the configured fallback chain is not used when the primary model returns a 429 rate-limit error. The run fails with LiveSessionModelSwitchError instead of falling back. The main agent loop correctly catches LiveSessionModelSwitchError and continues with the next model. But subagent/embedded runs do not — the error propagates as fatal and the run dies.

Fix Plan

To fix the issue, we need to add the missing error handling for LiveSessionModelSwitchError in the subagent/embedded run path.

Step-by-Step Solution:

• Modify the agent-runner.runtime-*.js file to include the error handling for subagent runs.
• Add a try-catch block around the subagent run code to catch LiveSessionModelSwitchError.
• Update the subagent run code to continue with the next model in the fallback chain when a LiveSessionModelSwitchError occurs.

Example Code:

// Subagent run code
try {
    // Existing subagent run code
} catch (err) {
    if (err instanceof LiveSessionModelSwitchError) {
        params.followupRun.run.provider = err.provider;
        params.followupRun.run.model = err.model;
        continue; // Continue with the next model in the fallback chain
    } else {
        throw err; // Rethrow other errors
    }
}

Verification

To verify the fix, test the subagent run with a primary model that returns a 429 rate-limit error. The subagent should now correctly fall back to the next model in the chain and continue running.

Extra Tips

• Make sure to update the agent-runner.runtime-*.js file in all relevant environments.
• Consider adding logging to track when the fallback chain is used to help with debugging and monitoring.