v1.3.11

v0.28.7 Release Notes

Released


v0.28.6 -> v0.28.7

Summary


Streamlines TUI output by collapsing repeated tool-call activity into a single summary line and surfacing cache-hit rates in session logs. Closes a gap where non-deterministic tool ordering silently broke prompt-cache reuse across requests. Planning triggers now cover analysis and research tasks, not just deep multi-step ones.
翻譯
精簡 TUI 輸出,將重複的工具呼叫活動收合成單行摘要,並在 session 紀錄中顯示快取命中率。修正工具排序不穩定導致提示詞快取靜默失效的問題。規劃觸發條件現已涵蓋分析與研究類任務,不僅限於深度多步驟情境。

Changes


FEAT

  • Collapse repeated tool-call activity into a single summary line in the TUI (@pardnchiu) [f687912]
  • Surface cache-hit percentage in session done-event logs (@pardnchiu) [f687912]

翻譯
  • 將重複的工具呼叫活動於 TUI 收合為單行摘要
  • 於 session 完成事件紀錄中顯示快取命中率


FIX

  • Sort tool definitions before assembly so ordering no longer varies across requests, restoring prompt-cache hits (@pardnchiu) [f687912]

翻譯
  • 組裝前排序工具定義,避免順序在每次請求間變動,恢復提示詞快取命中


UPDATE

  • Reposition the plan checklist and thinking-block rendering in the TUI (@pardnchiu) [f687912]
  • Broaden the write_todo trigger to cover analysis/research tasks without an active skill (@pardnchiu) [f687912]

翻譯
  • 調整 TUI 中 plan checklist 與 thinking block 的呈現順序
  • 擴大 write_todo 觸發範圍,涵蓋無啟用 skill 時的分析/研究類任務


CHORE

  • Refresh sitemap lastmod dates for the current build (@pardnchiu) [f687912]

翻譯
  • 更新本次建置的 sitemap lastmod 日期


Scope


  • internal/runtime/tui/ — FEAT, UPDATE (handlerExec.go, init.go, view.go, viewRender.go)
  • internal/session/log/ — FEAT (format.go)
  • internal/toolAdapter/ — FIX (api/adapter.go, mcp/mcp.go, script/translator.go)
  • configs/prompts/reasoning/, internal/tools/guide/ — UPDATE (write_todo.md, reasoningGuide.go)
  • worker/public/ — CHORE (sitemap.xml)



Generated by SKILL