A standalone session "cleaner & manager" for DSH — recoverable trash-bin delete, archive, cache cleanup (no plugin needed) 🐋

[BISTU-guheihei]

Hi everyone! I built a standalone session management tool for DeepSeek Harness — it's a native app, not a plugin, and that's exactly the point.

Why standalone?

DSH plugins extend the UI; this tool manages the DSH data itself — they complement each other. As a standalone app it doesn't rely on the DSH plugin system: no installation into DSH, no modification of DSH, no Python needed. It's a dedicated session operations tool: open it when you want, close it when you don't, and script it for batch maintenance.

What it does

• 🗑 Trash-bin delete — sessions go to a recycle bin first and can be restored anytime. Only an explicit purge actually frees space. Most in-DSH deletion is irreversible; this one gives you a second chance.
• 🗂 Archive / unarchive — reuse DSH's own archivedSessionIds, hide or restore sessions from the active list.
• 🔄 Live auto-watch — refreshes every 3 seconds, so new/deleted sessions in DSH reflect immediately.
• 📊 Session stats — turns, steps, LLM/tool time, token usage, context pressure bar.
• ✓ Deep cache cleanup — deleting a session also cleans workspace.json and session_projcache.json entries, then verifies no residue is left.
• 🛡 Safe — current session always protected, dangerous ops need double confirmation, archived sessions never lose data.
• 🎨 Dark whale-girl themed GUI + a full CLI for scripting (delete-old, clean --yes, ...).

📦 Download (no Python needed)

Platform	Artifact
Windows	DSH_SessionManager.exe — double-click to run
Linux	ELF binary — chmod +x && ./DSH_SessionManager

Or run from source: python3 manage_sessions.py (Python 3.8+, tkinter).

🔗

• Repo: https://github.com/BISTU-guheihei/DSH-SessionManager
• MIT licensed; whale-girl icons CC BY-NC-SA 4.0 (non-commercial)

Feedback and feature requests welcome! 🐋

---

🐋 中文介绍

大家好！我做了个 DSH 会话管理的独立工具——插件增强的是 DSH 的界面，而这个工具管理的是 DSH 的数据本身，两者互补。

• 独立的会话运维工具：与插件生态互补——插件增强 DSH 界面，这个工具管理 DSH 的数据本身。无需安装、无需 Python、独立运行，随时打开随时关闭，适合批量清理与脚本化运维；
• 🗑 回收站删除：删除先进回收站，随时可恢复；只有明确"彻底删除"才释放空间——很多界面内的删除是不可逆的，这个给你反悔的机会；
• 🗂 归档/恢复：复用 DSH 自身的 archivedSessionIds 机制；
• 🔄 实时监视：3 秒自动刷新，DSH 里的新会话/删除实时反映；
• 📊 会话统计：轮数、步骤、耗时、Token、上下文压力；
• ✓ 深度清理：删会话时同步清理 workspace.json / session_projcache.json 并复查无残留；
• 🛡 安全：当前会话永不删、危险操作双重确认、归档不丢数据；
• 🎨 深色鲸鱼娘主题 GUI + 完整命令行（delete-old、clean --yes 可脚本化）。

📦 下载即用（无需 Python）：Windows exe / Linux ELF，仓库 Release 里都有。

🔗 仓库：https://github.com/BISTU-guheihei/DSH-SessionManager

欢迎反馈和建议！🐋

[BISTU-guheihei]

有任何功能需求或反馈欢迎留言，我会持续更新！🐋

[denial123789]

This fills a real gap in the current UI, especially recoverable trash and retention. I compared the proposed behavior with the rc.2 persistence contracts and found a boundary worth making explicit:

• Live archive can use the official workspace.archiveSession RPC; it is serialized by the workspace registry and retains the log.
• Physical trash/restore should require every process sharing the Session and storage roots to be stopped. The JSONL backend documents one live writer per Session; “not the current Session” does not prove quiescence.
• session_projcache is a derived, identity-bound fold shortcut and self-heals from the log, so direct cache edits are not required for authoritative deletion.
• The normalized project directory is intentionally lossy. It should not be decoded back into cwd; validate the immutable Session header instead.
• A trash target should create a unique, no-overwrite generation. Reusing .trash/<sessionId> risks erasing the previous recovery point.
• If writer detection is uncertain or errors, mutation should fail closed. A whole-root snapshot should precede the first multi-file change.

I wrote up a source-pinned protocol with staged cross-device moves, crash journaling, restore collision rules, and 18 acceptance checks: https://sandbaseai.github.io/deepseek-harness-handbook/session-archive-trash-delete.html

Disclosure: I maintain the community DeepSeek Harness Handbook linked above. It is independent of DeepSeek AI.