claude-code - 💡(How to fix) Fix Resumed sessions with many images fail with unrecoverable dimension error [2 comments, 2 participants]

When a session accumulates many images (e.g. screenshots pasted during a long coding thread) and is later resumed via the SDK's resume option, the API rejects the request with a 400 error about image dimensions exceeding the limit for many-image requests. The SDK catches this error but surfaces it as the model's response text with no recovery path - particularly in non-interactive (SDK) mode where the only suggestion is "Start a new session with fewer images." I wrote a workaround that detects this error, strips the oldest image from the session JSONL, and retries. It peels one image per attempt until the API accepts. This confirms the session is otherwise valid - only the bulk image submission is the problem.

TL;DR

Modify the SDK to replay the conversation incrementally during resume, feeding messages to the API one by one, to avoid triggering the stricter per-image dimension limit for many-image requests.

Guidance

• Identify the point at which the SDK reconstructs the conversation from the JSONL and sends it to the API in one shot during resume.
• Modify the SDK to iterate through the JSONL and send messages to the API incrementally, mimicking the original turn-by-turn ingestion.
• Implement a check to ensure that the API accepts each turn during the replay, and handle any errors that may occur.
• Consider incorporating the provided workaround as a fallback, which strips the oldest image from the session JSONL and retries, to ensure session validity.

Example

# Pseudocode example of incremental replay
def resume_session(session_id):
    jsonl = load_jsonl(session_id)
    for message in jsonl:
        # Send message to API incrementally
        response = api.send_message(message)
        if response.error:
            # Handle error, potentially using the workaround
            handle_error(response.error)

Notes

The provided workaround confirms that the session is otherwise valid, and only the bulk image submission is the problem. However, this approach may not be suitable for all scenarios, and a more robust solution would be to modify the SDK to replay the conversation incrementally.

Recommendation

Apply the workaround, as it provides a functional solution to the issue, although it may not be the most efficient or scalable approach. A more permanent fix would involve modifying the SDK to replay the conversation incrementally, which would require further development and testing.