质量管线优化
三级质量模式 (quality_mode)
| 模式 |
质量步骤 |
适用场景 |
full |
verify → business-test → review → test-gen → test |
有 REQ-*.md 的正式 phase |
standard (默认) |
verify → review → test (test-gen 按条件) |
大多数场景 |
quick |
verify → CLI-review |
快速迭代、小修改 |
条件执行与轻量替代
- business-test 改为
full 模式专属(与 quality-test 40% 重叠,standard/quick 省略)
- test-gen 改为条件执行(standard 模式仅在覆盖率 < 80% 时触发)
- quick 模式 review 改用
maestro delegate --role review CLI 轻量替代
轻量重试循环
- 引入
passed_gates[] 追踪已通过的质量门
- post-test 失败不再重插整条管线,仅重跑 verify + 受影响的门
- 按修改文件与门检查范围的交集判断是否需要重跑
-y 自动模式
Ralph 支持 -y
$maestro-ralph "intent" -y — decision 节点自动评估并继续(不 STOP)
post-debug-escalate 始终 STOP(安全阀)
- 失败自动重试一次后跳过
传播链
ralph -y → session.auto = true
→ wave CSV: $maestro-ralph-execute -y "$skill"
→ ralph-execute 按映射表附加 -y 到目标 skill
统一 -y 参数
所有命令 auto 参数统一为 -y,替换原有的 --auto、--auto-yes:
| 命令 |
-y 效果 |
| maestro-init |
跳过交互提问 |
| maestro-analyze |
跳过交互 scoping |
| maestro-brainstorm |
跳过交互提问 |
| maestro-roadmap |
跳过交互选择 |
| maestro-plan |
跳过确认和澄清 |
| maestro-execute |
跳过确认,blocked 自动继续 |
| quality-business-test |
跳过计划确认 |
| quality-test |
自动触发 gap-fix loop |
| quality-retrospective |
自动接受路由推荐 |
| maestro-milestone-complete |
跳过 knowledge promotion 交互 |
修复
- ralph-execute 字段名
commands[] → steps[] 对齐 ralph SKILL.md
- decision 节点命名统一为
post-biz-test