CrewAI 1.15.24
CrewAI 1.15.24 introduces evaluation enhancements including model-swapping and brief generation, Oracle integrations, and experimental features for job lifecycles and reply queues.
原始内容为英文；当前页面提供中文导航与来源说明，具体事实请以原文为准。
为什么值得关注
CrewAI 1.15.24 brings several feature updates, bug fixes, and refactoring efforts. Key features include adding crewai eval to generate markdown briefs when executed by an agent, support for crewai eval --models alongside LLM overlays to swap models and roles, Oracle integrations, and background reply queueing. It also adds experimental capabilities for turn/reply identities and a job lifecycle/runner. Fixes address CLI train filename errors, start step re-runs based on listened events, output exposure of human feedback reviews, non-zero exit codes (code 1) for failed crewai eval gates, tracing grant exporter usage reporting, and multiple package dependency advisories. Internal refactoring centralizes context windows and moves message summarization into SummarizeMessages .
可执行摘要
CrewAI v1.15.24 introduces crewai eval CLI tooling with markdown reports and model swapping, experimental job lifecycle runners, Oracle integrations, context window centralization, and dependency security updates.
- Agent 实用度
- 62/100
- 可信度
- 96%
- 机器格式
- JSON + Markdown
开发者应核对什么
- Upgrade to CrewAI 1.15.24 to apply fixes for multiple dependency advisories.
- Update CI/CD pipelines relying on crewai eval to handle exit code 1 when evaluation gates fail.
- Test Oracle integrations or experimental job lifecycle runners if adopting these components.