Session Recall

Session Recall 會讀取你自己的 ~\.copilot\session-store.db,建立一個本機可搜尋的 session 索引,讓 Copilot CLI 可以回想相似的過去對話、匯出 context pack,並產生關聯圖。

Session Recall reads your own ~\.copilot\session-store.db and builds a local, searchable session index so Copilot CLI can recall similar past conversations, export context packs, and generate a relationship graph.

所有資料都留在你的電腦上,不會由這個工具上傳。

Everything stays on your machine and is never uploaded by this tool.

DeepWiki

你也可以在 DeepWiki 上閱讀自動產生的 repo 說明與架構導覽:

You can also read the auto-generated repo documentation and architecture guide on DeepWiki:

DeepWiki: iouovonoi/session-recall

實測概覽 / Results at a glance

以下數字來自一個真實本機索引的參考測試。完整說明請看 session-recall/docs/validation.md

The numbers below come from a real local reference index. See session-recall/docs/validation.md for details.

指標 / Metric 數值 / Value
找到相關 session 時節省的 context / Context saved when a relevant session is found 60% to 97%
單次實測查詢 / One measured query 18,382 chars to 505 chars, 97.2% saved
3 筆 benchmark 命中率 / 3-query benchmark hit rate hit@1 = hit@5 = 66.7%
找不到相關內容時 / When nothing relevant exists 回報無匹配,不硬塞低信心結果 / Reports no match instead of forcing a low-confidence result

先看哪份文件?/ Which doc should I read?

文件 / File 適合對象 / Who it is for 內容 / What is inside
README.md 一般使用者 / General users 安裝步驟、隱私規則、常用指令 / Install steps, privacy rules, common commands
session-recall/REFERENCE.md 進階使用與排錯 / Power users and troubleshooting 完整功能、JSON 格式、FAQ、內部設計 / Full feature list, JSON shape, FAQ, internals
session-recall/docs/validation.md 想確認效果的人 / Anyone who wants proof it works 報告與 benchmark 數字說明 / Report and benchmark numbers explained
session-recall/SKILL.md Copilot CLI / agent 觸發規則與使用方式 / Trigger rules and agent-facing instructions

建議先讀 README;遇到問題或想看完整欄位時,再打開 session-recall/REFERENCE.md

Start with this README. Open session-recall/REFERENCE.md when you need troubleshooting details or the full field reference.

安裝 / Install

把 repo 裡的 session-recall 資料夾複製到:

Copy the session-recall folder in this repo to:

%USERPROFILE%\.copilot\skills\session-recall

重新啟動 Copilot CLI。第一次使用時,允許 agent 執行安裝,然後執行:

Restart Copilot CLI. On first use, approve the agent request, then run:

powershell -ExecutionPolicy Bypass -File "$env:USERPROFILE\.copilot\skills\session-recall\install.ps1"

可選參數:

Optional flags:

需求 / Need 參數 / Flag
更改預設匯入 200 筆 session 的限制 / Change the default 200-session import limit -InitialSyncLimit 500
只建立空索引,不匯入資料 / Create an empty index and skip import -SkipInitialSync
每 30 分鐘自動同步 / Auto-sync every 30 minutes -InstallScheduler -EveryMinutes 30

測量 context 節省量 / Measuring context savings

使用 --report 可以確認 recall 是否找到相關歷史,以及建議注入的 context 是否變小。

Use --report to check whether recall finds relevant history and whether the injected context becomes smaller.

powershell -ExecutionPolicy Bypass -File "$env:USERPROFILE\.copilot\skills\session-recall\scripts\session-recall.ps1" sync-copilot --limit 50 --compare "session recall validation" --compare-limit 5 --report

如何解讀數字、哪些說法適合公開引用,請看 session-recall/docs/validation.md

For how to read the numbers and what is safe to claim publicly, see session-recall/docs/validation.md.

隱私 / Privacy

可以分享 / Share this 不要分享 / Never share
skill 原始檔:.md, .ps1, .py / Skill source files ~\SessionRecall
文件 / Documentation ~\.copilot\session-store.db
已移除個資的截圖 / Screenshots with personal content removed ~\.copilot\session-state or other memory stores

預設產生的資料都在 ~\SessionRecall,這個工具不會把資料上傳。

Generated data stays under ~\SessionRecall by default. This tool does not upload it.