[CopilotClient] aiohttp "Chunk too big" error on large responses in microsoft-agents-copilotstudio-client

[LeoBERTAU]

Describe the bug

I am using microsoft-agents-copilotstudio-client (v0.6.1). When the Copilot agent returns a large response (e.g., a long summary > 64KB), the underlying aiohttp client crashes with a Chunk to big error.

The library does not expose a way to configure the aiohttp.ClientSession or increase the readuntil limit, which defaults to 64KB. This is often insufficient for Generative AI responses sent as single-line JSON or SSE events.

To Reproduce

1. Initialize CopilotClient.
2. Send a query that generates a large text response (> 64KB).
3. The client raises aiohttp.streams.LineTooLong (or similar).

Expected behavior

The client should allow configuring the session limit or default to a higher limit (e.g. 10MB) to handle verbose LLM outputs.

Workaround

I am currently using this monkeypatch to fix it:

import aiohttp.streams

_original_init = aiohttp.streams.StreamReader.__init__

# Patch default limit to 10MB
def _patched_init(self, *args, **kwargs):
    if 'limit' not in kwargs or kwargs['limit'] == 65536:
        kwargs['limit'] = 10 * 1024 * 1024
    _original_init(self, *args, **kwargs)


aiohttp.streams.StreamReader.__init__ = _patched_init

This will alert the maintainers specifically for the Python SDK. Good catch on finding the specific Python repo