OpenAI Agents API adds auto review, background compaction and automatic system-prompt caching

An OpenAI engineer on the Agents API team listed the week's changes; the OpenAI Developers account posted a pointer to the same list. According to that post:

• Auto review. As in Codex, commands are evaluated for potentially bad behavior before they run. OpenAI's API changelog says it is available for eligible Agents API projects: set the session's review field to "auto_review" and a model approves or denies supported tool and permission requests based on conversation intent and risk. The changelog notes limitations and additional token billing, described in the automatic review guide.
• Async compaction. Compaction can happen in the background, which is meant to speed up sessions. OpenAI's changelog says it now runs in the background by default; set compaction.async to false for blocking compaction. Background compaction may increase token usage.
• Time awareness. An experimental feature that gives the model a sense of time passing, so it can act on requests such as "timebox CI fixes to 10 minutes". Per the changelog it is opt-in and off by default: set experimental.time_awareness to true to give the model periodic current-time reminders.
• Tool updates. A session's tools can now be updated at any time. The changelog says an update replaces the full tool list; function tools and MCP servers change on the next model call, built-in tools by the next turn.
• Prompt caching. System prompts are now cached automatically across all sessions, which OpenAI says saves money and time.