改了什么
- 用
later设下的复查到期时,派给 Agent 的资料里context.timers[].events带上设复查那一轮在做的事件编号。业务系统用自己的编号发事件、Agent 设了复查,复查那一轮做出来的结果也能对回原来的编号(#333)。 - 飞书通道:飞书那边还记着旧的事件连接、拒绝新的卡片点击监听时(常见于刚重启),日志里第一次就用中文说清“卡片点击暂时收不到”和该查什么,之后每分钟重试一次,不再每 10 秒刷一遍英文报错;停中台时不再干等(#337,修 #335)。
- 交付插件的审查改成两级:每条发现标“挡上线”或“建议”,只有挡上线的退回改动;连续两轮都有挡上线就交给负责人定,可信账号写
delivery-owner/<PR 编号>状态放行;审查读分支相对 main 的最终改动,而不是逐个提交;只用文档回应挡上线的发现时,下一轮拿整份改动核对(#342、#343、#344、#345)。
感谢 argszero。
What's changed
- When a check set with
latercomes due,context.timers[].eventsin the job carries the ids of the events the round that set it was working on, so a source that posted under its own id can tie the check's result back to it (#333). - Feishu channel: when Feishu still holds an old event connection and refuses the card-click listener (common right after a restart), the log says so once, in plain words, with what to check; it then retries once a minute instead of logging the raw error every 10 seconds, and stopping no longer waits out the delay (#337, fixes #335).
- Delivery review now has two levels: each finding is marked blocking or a suggestion, and only blocking ones send the change back. After two rounds in a row with blocking findings it goes to the owner, who can let it through with a
delivery-owner/<PR number>status from a trusted account. Review reads the branch's final diff against main rather than each commit, and a blocking finding answered with docs only is checked against the whole change in the next round (#342, #343, #344, #345).
Thanks to argszero.