OpenAI Agents SDK JavaScript v0.17.1
OpenAI Agents SDK JavaScript v0.17.1 adds server-wide MCP guardrails, custom output guardrail blocked messages, Docker sandbox container labels, and image results in web search tools, alongside a critical security fix for Git argument injection in sandboxes and numerous state-handling bug fixes.
Why this signal matters
OpenAI Agents SDK JavaScript release v0.17.1 includes several new features and significant bug fixes. New capabilities include server-wide MCP tool guardrails, custom messaging when output guardrails block responses, labeling support for Docker sandbox containers, and image result support in web search tools. Key security and correctness fixes address Git repository argument injection in sandboxes, preservation of serialized approval ownership across resumes and agent handoffs, recovery of failed resumed session writes, deduplicated MCP server lifecycles, and thread-safe PTY and trace handling.
Actionable summary
SDK release v0.17.1 introduces server-wide MCP tool guardrails, custom blocked output messages, and Docker sandbox labels. It fixes a sandbox Git argument injection vulnerability, resolves session recovery and approval handoff state persistence bugs, and handles trace exports and PTY cleanup safely.
- Agent usefulness
- 82/100
- Confidence
- 96%
- Canonical data
- JSON + Markdown
What builders should check
- Upgrade openai-agents to v0.17.1, especially if using sandbox environments exposed to untrusted Git URLs to remediate argument injection (PR 1840).
- Review MCP tool configurations if relying on global enforcement, taking advantage of the new server-wide MCP guardrails.
- Verify approval resume workflows and agent handoffs if encountering lost state or dropped session writes in previous versions.