Pin openai-agents to stable release once reasoning-content replay is released

[dittops]

Context

We upgraded openai-agents from 0.3.3 to a git install from main to get the reasoning-content replay hook (should_replay_reasoning_content) from openai/openai-agents-python#2670.

This hook is needed because our LLM provider (Moonshot AI via Bud Foundry) requires reasoning_content on assistant messages in multi-turn conversations. Without it, the provider fails. The stock SDK only replayed reasoning content for DeepSeek/Claude model names, dropping it for all others.

What to do

Once openai-agents publishes a stable release containing PR #2670 (merged 2026-03-20), update backend/requirements/default.txt:

-# TODO: Pin to release once reasoning-content replay hook is in a stable release
-# Tracking: https://github.com/BudEcosystem/onyx/issues/XXX
-# PR: https://github.com/openai/openai-agents-python/pull/2670
-openai-agents @ git+https://github.com/openai/openai-agents-python.git@main
+openai-agents==<new_version>

Also remove git from the Dockerfile apt-get install if no other dependency needs it.

Files changed

• backend/requirements/default.txt — git dependency
• backend/Dockerfile — added git to apt-get for git+https pip install
• backend/onyx/agents/bud_agent/agent_context.py — uses should_replay_reasoning_content hook