feat(multi-review): add regression-test reviewer persona

[Svtter]

Summary

Adds a built-in regression-test reviewer persona to multi-review, so every multi-review run automatically audits PRs for missing regression tests alongside quality, security, performance, and architecture checks.

Changes

multi-review/reviewers/regression-test.yaml — new persona that:

• Classifies PR type (BUGFIX, BEHAVIOR_CHANGE, NEW_FEATURE, CHORE)
• Only flags missing tests for BUGFIX and BEHAVIOR_CHANGE PRs
• Skips NEW_FEATURE and CHORE with "无需回归测试"
• Uses severity levels: CRITICAL / MEDIUM / LOW
• Follows the same output format as other reviewers (decision → summary → blocking → suggestions)

multi-review/src/reviewers.ts — 2 changes:

• Added regression-test.yaml to loadBuiltInReviewers file list
• Added regression-test:1 to DEFAULT_TEAM (now 5 reviewers by default)

Behavior change

Default team changes from quality:1,security:1,performance:1,architecture:1 (4 reviewers) to quality:1,security:1,performance:1,architecture:1,regression-test:1 (5 reviewers). Users who explicitly set default-team are unaffected.

[github-actions]

无需回归测试

PR 类型：NEW_FEATURE

该 PR 为 multi-review 系统新增了一个 regression-test 内置评审人格（reviewer persona），主要变更包括：

• 新增 multi-review/reviewers/regression-test.yaml 配置文件（57 行），定义了回归测试评审的逻辑和输出格式
• 在 multi-review/src/reviewers.ts 中注册该评审人格并加入默认团队
• 更新了构建产物 multi-review/dist/index.cjs

这是纯新增功能（添加新的评审维度），虽然默认团队从 4 人变为 5 人（包含了新人格），但这属于新增功能的自然扩展而非对既有逻辑的修改。PR 未修改现有 reviewer 的任何行为逻辑，也未修复任何 bug，因此不需要回归测试。

opencode session  |  github run

[github-actions]

最终决策：可合并 / CAN MERGE

四名评审人一致认为 PR 改动量小、遵循现有模式、无功能性错误。

阻塞项：无

建议项：

1. 检查仓库中是否存在检测 DEFAULT_TEAM 精确内容或内建评审人列表的测试用例，若有则需同步更新，否则可能造成测试失败（来自 quality 评审人）

---

📋 各 Reviewer 详细审查结果

quality

可合并 / CAN MERGE

该 PR 新增了一个 regression-test 评审人角色，改动量小且模式高度一致，无功能性错误。

• 在 DEFAULT_TEAM 字符串末尾追加 regression-test:1，格式与现有条目一致
• 在 loadBuiltInReviewers 的数组末尾追加 "regression-test.yaml"，加载机制与现有评审人相同
• regression-test.yaml 文件结构参照了现有评审人模板，prompt 指令清晰完整
• 源码 (src/reviewers.ts) 和构建产物 (dist/index.cjs) 同时修改且一致

阻塞项：无

建议项：

1. 建议确认仓库中是否存在检查 DEFAULT_TEAM 精确内容或内建评审人列表的测试用例，若有则需同步更新，否则可能造成测试失败

security

安全无虞 / SECURE

本次 PR 仅向 DEFAULT_TEAM 添加 regression-test:1 评审角色、注册新的 regression-test.yaml 评审配置文件。代码量小，功能单一。

阻塞项：无

建议项：无

performance

性能良好 / GOOD

该 PR 仅新增了一个 regression-test 评审角色，改动量极小：

• DEFAULT_TEAM 字符串增加一项，parseTeam 解析时多处理一个键值对，复杂度不变
• loadBuiltInReviewers 多加载一个 YAML 文件（57 行），增加一次同步文件读取和 YAML 解析，开销可忽略
• Map 中多一个条目，内存增量极微

未引入任何算法复杂度变化、内存泄漏、数据库查询、并发问题或资源瓶颈。

阻塞项：无

建议项：无

architecture

架构合理 / SOUND

该 PR 新增一个 regression-test 评审角色，改动量极小且完全遵循现有架构模式。新 reviewer 的 YAML 定义与已有 quality.yaml、security.yaml 等保持一致的 name + prompt 结构，src/reviewers.ts 仅在数组和默认团队字符串中各加一项，改动集中于单一模块，无跨模块耦合。

阻塞项：无

建议项：无

[Svitter]

Review: PR #212 — feat(multi-review): add regression-test reviewer persona

Rank: P2 (needs maintainer sign-off) — feat, agent

Clean PR. Three files, well-structured prompt, follows existing reviewer conventions. The regression-test persona is a sensible addition to multi-review.

Gate

Expanding DEFAULT_TEAM from 4 to 5 reviewers at multi-review/src/reviewers.ts:10 is a default-behaviour change. Every multi-review run that does not explicitly set default-team now gets an additional AI reviewer call, increasing runtime and API cost. The PR body acknowledges this — just needs the maintainer to confirm it's intentional before merge. This alone caps the PR at P2 per the "no default-behaviour changes without explicit maintainer sign-off" convention.

Findings

1. multi-review/src/reviewers.ts:25 — missing test coverage. No test verifies that regression-test.yaml successfully loads and produces a valid PersonaYAML with the expected name field. The existing reviewers.test.ts uses MULTI_REVIEW_DEFAULT_TEAM = "quality:1" which loads the new persona into the map but never exercises it as a reviewer. Consider adding a test that loads regression-test:1 and asserts the reviewer name and that the prompt is non-empty.

2. multi-review/README.md:22 — stale reviewer list. The documented built-in reviewer personas list (quality, security, performance, architecture) is now incomplete. Should include regression-test.

What's good

• The prompt is well-designed: clear PR-type classification (BUGFIX/BEHAVIOR_CHANGE/NEW_FEATURE/CHORE), correct scoping (flags missing tests only for BUGFIX/BEHAVIOR_CHANGE), and follows the same blocking/suggestions output format as other reviewers.
• The dist/index.cjs rebuilt output matches the source changes exactly.
• Decision vocabulary (无需回归测试/回归测试完整/缺少回归测试) is domain-appropriate, consistent with how other reviewers have their own decision sets.

Thanks @Svtter for the contribution.