@openai/agents-extensions@0.19.0
@openai/agents-extensions@0.19.0 updates sandbox requirements, sanitizes sensitive tracing and authentication data, and refines AI SDK error handling.
Why this signal matters
The 0.19.0 release of @openai/agents-extensions introduces a minor change requiring Daytona SDK ^0.201.0 to include patched OpenTelemetry dependencies while maintaining polling-based sandbox lifecycle behavior. Numerous patch-level fixes focus on tracing and credential safety: treating empty CODEX API KEY values as unconfigured to allow OPENAI API KEY fallback, honoring sensitive data tracing policies across Codex spans, omitting reasoning outputs from default OpenAI trace exports, redacting MCP authentication headers from AI SDK UI tool search outputs, and omitting raw provider bodies and headers from error traces. Other improvements include sandbox concurrency management during directory copies, archive limit enforcement, and Blaxel prefix trimming fixes alongside dependency updates to @openai/agents-core@0.19.0.
Actionable summary
OpenAI Agents Extensions 0.19.0 bumps Daytona SDK to ^0.201.0, sanitizes MCP auth headers and raw provider error traces, omits default reasoning exports, and aligns with @openai/agents-core@0.19.0.
- Agent usefulness
- 77/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade Daytona SDK dependency to ^0.201.0 if using Daytona sandbox extensions.
- Verify telemetry, trace exports, and error monitoring setups accommodate redacted provider error details and omitted default reasoning outputs.
- Update companion package @openai/agents-core to version 0.19.0.