Description

In getReplyFromConfig (reply pipeline), applyMediaUnderstandingIfNeeded is called without a try/catch wrapper. While the internal layers have their own error handling, an unexpected error (e.g., from cache.cleanup() in the finally block, or a bug in finalizeInboundContext) could crash the entire reply pipeline.

Current Behavior

If applyMediaUnderstandingIfNeeded throws an unhandled error, the entire getReplyFromConfig call fails and the user gets no response at all — not even a text-only response ignoring the media.

Suggested Fix

Wrap applyMediaUnderstandingIfNeeded in try/catch and fall back to processing the message without media understanding. A degraded response (text-only, no transcription) is better than no response.

try {
  await applyMediaUnderstandingIfNeeded({ ctx, cfg, agentDir, activeModel });
} catch (err) {
  logger.warn('Media understanding failed, continuing without it', err);
}

Environment

• OpenClaw 2026.4.2