v0.3.3
[0.3.3] - 2026-04-10
Added / 新增
- Regenerate option: Press
[r]in the confirmation menu to skip cache and re-query the LLM - 重新生成选项:在确认菜单按
[r]跳过缓存重新请求 LLM,无需手动清缓存 - Command sanitization: Auto-strips leading garbage characters (
:,$,>) from LLM output, improving compatibility with smaller models - 命令清洗:自动去除 LLM 输出的前导垃圾字符(
:,$,>),提升中小模型兼容性 - No-op command rejection: Empty commands,
:,true, andnoopare rejected instead of being cached or executed - 空命令拦截:空命令、
:、true、noop等无效命令被拒绝,不再缓存或执行
Improved / 优化
- Cache key includes model: Switching LLM models auto-invalidates stale cache entries (key now includes backend:model)
- 缓存隔离:缓存 key 加入 model 标识,切换模型后旧缓存自动失效
- Cache key includes package manager: Context-sensitive queries (e.g. "build project") return correct results per project type
- 上下文感知缓存:缓存 key 加入包管理器信息,相同查询在不同项目类型下返回正确结果
- Default cache TTL reduced from 168h to 48h: Shorter window for LLM errors to persist
- 默认缓存 TTL 从 168h 降至 48h:缩短错误结果的影响窗口
- Total tests: 400 (355 unit + 45 integration)
- 测试总数:400(355 单元 + 45 集成)
Full Changelog: v0.3.2...v0.3.3