[Release] openai/openai-agents-python v0.17.1: v0.17.1

OpenAI Agents Python v0.17.1 is a maintenance release focused on bug fixes and stability improvements across sandboxes, tracing, sessions, realtime agents, and core modules. Key fixes address sandbox archive extraction, session data corruption handling, realtime tool approvals, chat completion validation, and various edge cases in handoff logic and schema handling. The release includes 50+ targeted fixes to improve reliability and data integrity across the framework.

Key Points

• •Sandbox improvements: Fixed archive extraction limits, git repo subpath validation, and sandbox provider error details reporting
• •Tracing enhancements: Made tracing shutdown best-effort on process exit and improved BatchTraceProcessor resilience to exporter errors
• •Session stability: Fixed corrupt item handling in MongoDB and Redis sessions, preserved metadata timestamps, and maintained created_at across writes
• •Realtime agent fixes: Scoped tool approvals by qualified key, fixed unknown tool call output, and improved audio input/output handling
• •Chat completions validation: Made response-feature validation opt-in and rejected unsupported features like server state and reusable prompts
• •Handoff improvements: Preserved nested handoff history content, fixed async callable handling, and replaced assertions with proper UserError exceptions
• •Schema handling: Fixed nested $ref expansion, aligned generic dict output schemas, and isolated MCP strict schema conversion
• •Data preservation: Maintained request usage entries, output token details, tool guardrail results, and response IDs across operations
• •Error handling: Improved guardrail exception cleanup, validated model retry backoff settings, and added proper error handling for edge cases
• •Tool management: Fixed tool trimmer allowlist handling and preserved HandoffInputData.input_items in tool removal operations

Workflow Diagram

Concepts