# Changelog ## 0.1.4 — 2026-09-18 ## 简体中文 - `/` 菜单为 `/review`、`/review-status`、`/review-cancel` 补上图标和中文标题,描述继续跟随宿主语言。 - 将 npm 包描述改为与 GitHub 仓库一致的中英双语。 ## English - Add slash-menu icons and localized titles for `/review`, `/review-status`, and `/review-cancel`; descriptions still follow the host locale. - Make the npm package description bilingual to match the GitHub About text. ## 0.1.3 — 2026-09-18 ## 简体中文 - 将 peer 与开发依赖对齐官方 DSH `0.1.6-alpha.2`,并适配 `SubagentRuntime` 构造函数;已在该宿主上完成离线回归与安装包端到端验证。 ## English - Align peer and development dependencies with official DSH `0.1.6-alpha.2`, update the `SubagentRuntime` constructor, and verify offline regression plus packaged host integration on that host. ## 0.1.2 — 2026-09-16 ## 简体中文 - 将 peer 与开发依赖对齐官方 DSH `0.1.6-alpha.1`,并在该宿主上完成离线回归与安装包端到端验证。 ## English - Align peer and development dependencies with official DSH `0.1.6-alpha.1`, and verify offline regression plus packaged host integration on that host. ## 0.1.1 — 2026-09-15 ## 简体中文 - 增加 GitHub Actions:推送正式版本标签后,通过 npm Trusted Publishing 发布,并同步中英 GitHub Release 说明。 ## English - Add GitHub Actions so a version tag publishes via npm Trusted Publishing and syncs bilingual GitHub Release notes. ## 0.1.0 — 2026-09-13 ## 简体中文 - 在主会话使用 `/review` 发起原生独立 Agent 审查,支持基准分支、未提交更改、指定提交及自定义要求;报告返回当前会话,可查看最近结果或取消运行。 - 保留固定版本的 Codex 英文审查规范,报告语言跟随宿主中英文偏好,并提供双语使用说明。 - 支持 DSH `0.1.5-rc.2`,需要 Node.js ≥ `22.19.0`、Git、原生 spawn 与问题回答器。未保存语言偏好时默认中文;审查范围覆盖、发现准确性和输出语言仍取决于模型表现。 ## English - Start a native independent-agent review with `/review` in the main conversation. Choose a base branch, uncommitted changes, a commit, or custom instructions; receive the report in the same conversation, recover the latest result, or cancel a run. - Preserve the original English Codex rubric from a fixed source version, follow the host's Chinese or English preference for reports, and provide bilingual documentation. - Supports DSH `0.1.5-rc.2` and requires Node.js ≥ `22.19.0`, Git, native spawn, and a question answerer. Chinese is the default without a saved locale preference; review coverage, finding accuracy, and output language still depend on model behavior.