OpenAI Agents SDK Python v0.22.1

OpenAI released version 0.22.1 of the Agents SDK for Python. Key feature additions include support for image results in web search tools, customizable output guardrail blocked messages, server-wide guardrails for MCP tools, configurable Unix-local environment isolation and Docker container labeling in the sandbox, and exposed streaming transcription options for voice workflows. The release includes substantial bug fixes across several domains: - Core: Enforces fail-closed handling for empty tool arguments, preserves variadic tuple annotations/Field constraints, rejects kwargs key collisions with named tool parameters, fixes stream span drops, and closes model providers created by Runner. - Sessions: Hardens recovery for failed resumed writes, improves concurrent write handling during compaction, and prevents resuming runs where accepted terminal output failed to persist. - Voice & Realtime: Adds support for custom OpenAI TTS voice IDs, enforces redaction of STT prompts and TTS instructions in traces, and corrects cancellation and deadline behaviors in streamed audio. - Sandbox & MCP: Resolves unbounded event loop I/O for UnixLocal workspaces, handles PTY cleanup and cancellation, and prevents tool caching mutation by deep-copying cached MCP tools.

• Upgrade openai-agents to v0.22.1 via standard package management.
• Verify tool definition signatures if utilizing kwargs, variadic tuples, or empty argument payloads to ensure compatibility with stricter validation.
• Review MCP tool definitions to leverage new server-wide guardrails where appropriate.
• Test session resumption paths to benefit from enhanced write failure and compaction recovery.