InfoQ 中文T2·78 pts
- Trails replace PR timelines with multi-round review-agent scoring, allowing merges only when confidence is high, risk is low and all issues are cleared
- Review agents run in cloud sandboxes scoring code quality, security, slop and human-review needs, configurable via repo-level JSON and Gates
- Continuous memory uses two open-source tools to build a precomputed repository graph so agents reuse knowledge across sessions