DeepSeek Harness stores every conversation as an append-only event log, so there is no built-in "undo". dsh-message-editor brings back the three moves every chat deserves — 撤回 (recall), 编辑重发 (edit-and-resend) and 重新生成 (regenerate).

Recall / edit remove the target messages from the conversation view and the model context — that is exactly the effect you see. What stays untouched is the underlying durable transcript: it remains append-only, old events are never rewritten or deleted, and the plugin merely appends one valid replacement event (the same replace primitive the built-in compaction uses) to rewind the surface — so the log keeps a full audit trail of every rewind.

---

✨ Features

Action	Where	What happens
↩ 撤回 (recall)	hover any assistant reply, or the row under any user message	Removes the whole exchange round (the input and the agent's output, tool rows included) from both the model context and the conversation view; the input text is echoed into the composer so you can re-ask or re-edit immediately. A small transient notice marks the rewind and disappears once you keep typing.
✎ 编辑重发 (edit & re-send)	row under any user message	The old message and its reply are rewound and hidden. By default the conversation starts fresh (earlier messages are hidden too and excluded from context); the edited text is sent and the agent answers. A collapsed "original input" reference sits right under the new message — click to expand, configurable off.
↻ 重新生成 (regenerate)	hover any assistant reply	The reply (and everything after it) is rewound and hidden, then the original prompt is re-sent so the agent answers again.

Why it's different

• 🎯 Whole-round recall — one click removes the input and its output (including tool rows), not just a single bubble.
• 🖥️ Web + Desktop — the same plugin covers both surfaces of DeepSeek Harness.
• 🔒 Removed from view & context, not from the log — recalled/edited messages disappear from the conversation view and the model context, while the durable transcript is never rewritten or deleted; the plugin only appends valid, typed session events (the same replace primitive the built-in compaction uses), so the log keeps a full audit trail.
• 🧠 View ⇄ context in sync — the conversation view always reflects exactly what the agent sees.
• ⚡ Try in 30 seconds — the dynamic form installs in your current session with no rebuild.

---

🚀 Quick start

Requires DeepSeek Harness with the dsh CLI. Installs the plugin as a profile bundle and automatically rebuilds the Web client:

dsh plugin --profile web add dsh-message-editor

That's it — hover any assistant reply, or any user message, and use ↩ / ✎ / ↻.

---

📦 Installation

1. Profile bundle (recommended)

The package declares a dsh.bundle manifest, so it installs through the official plugin path into any profile:

dsh plugin --profile <name> add dsh-message-editor

It also shows up in dsh-market for one-click install from inside Settings.

2. npm package + composition (classic)

npm i dsh-message-editor

Add the package to the harness composition (cordis.yml of the app/deployment you use):

- name: 'dsh-message-editor'

The client half is picked up automatically from the package's dsh.client metadata and bundled into the Web client (a client-module rebuild happens automatically when the composition changes). The Host half registers the same-origin HTTP route /api/plugins/message-editor/* for the browser UI.

3. Dynamic plugin (current session — no install, no rebuild)

Use the dynamic entries shipped in the package. In the session where you want the feature:

1. Open the plugin editor and define a new plugin from lib/dynamic-host.js (Host half) and lib/dynamic-client.js (Client half).
2. Approve and run the Client half.
3. Done — hover any assistant reply, or any user message, and use ↩ / ✎ / ↻.

The dynamic host registers the same operations behind the package-private harness.handle RPC (messageEditor.recall / messageEditor.editAndResend / messageEditor.regenerate).

---

⚙️ Settings → General

Setting	Default	Description
编辑后显示原提问对照	on	A collapsed "original input" reference under the re-sent message showing the most recent replaced text (reference only — never sent to the model).
编辑后从新对话开始	on	After editing, hide earlier messages too so the conversation looks like a fresh start (the whole surface is rewound before re-sending).

---

🧠 How it works

 durable transcript (append-only)          model context & view
 ┌────────────────────────────────┐        ┌──────────────────┐
 │ ... target message             │        │  … target message │
 │     ↓ shadow span              │        │       ↓ rewind    │
 │ [target … last surface node]   │ ─────▶ │  (empty replace   │
 │     ↳ one replacement          │        │   = context cut)  │
 │       assistant/message (empty)│        └──────────────────┘
 │     ↳ optional original-input  │        agent.followup(new prompt)
 └────────────────────────────────┘        → next turn rebuilds request

1. Host core (lib/host-core.js, zero runtime imports) locates the target message in the session's live surface, computes the shadow span [message … last surface node], and appends one replacement assistant/message with an empty body — a valid surface node that derives to no model message, so the LLM context simply rewinds.
2. Edit / regenerate additionally call agent.followup(...) with the (new) prompt text; the agent's next turn builds its request from the rewound session.deriveMessages().
3. Client (lib/client.js) registers:
   • a user-actions conversation node under every user message (编辑 / 撤回 row with an inline editor); recall echoes the text into the composer,
   • the recall-marker node renderer: a notice row that injects CSS hiding every shadowed message row from the flow (view and model context stay in sync), plus the optional original-input comparison block,
   • the message-editor entry in the conversation.chat.assistant-actions strip (撤回 / 重新生成),
   • two preference toggles under Settings → General.

Two different layers are at play: the durable transcript (append-only; old events are never rewritten or deleted) and the model-visible surface (rewound by an appended replacement event). So the old events stay in the log as an audit trail — but they are synchronized out of both the model context and the visible conversation, and the view always reflects what the agent actually sees. Persistence, projections and the transcript remain consistent because the plugin only appends valid, typed session events.

---

⚠️ Requirements & limitations

• Only user messages can be edited; recall works on user and assistant messages. Tool results are shadowed along with the recalled range but are not themselves recall targets.
• The agent must be idle: while a reply is streaming you must stop it (⏹) before recalling or editing. The Host rejects with agent-busy otherwise.
• Recall/edit operate on the active model surface: a message that was already compacted away or previously recalled is rejected (target-shadowed).
• Regenerate re-sends only the text of the original prompt; prompts that carried images fall back to the text-only content.

---

🗺️ Roadmap

• Version timeline / reroll — browse and jump between past rewinds of a message
• Forked-session edit — edit a past message and continue in a branched session
• More locales beyond 简体中文 / English

---

🛠️ Development

# structure
lib/host-core.js       # transport-neutral host logic (no imports)
lib/index.js           # published Host: harness RPC + HTTP route
lib/client.js          # published Client (React via import, fetch transport)
lib/dynamic-host.js    # dynamic Host half (self-contained)
lib/dynamic-client.js  # dynamic Client half (self-contained)
cordis.patch.yml       # dsh.bundle profile patch layer

npm pack --dry-run     # verify the published file list
node --check lib/*.js  # syntax check

PRs and issues are welcome — see CONTRIBUTING (coming soon) and the issue tracker.

---

📚 Ecosystem

Listed on the dsh-plugin topic and installable from dsh-market. For a curated overview of the DeepSeek Harness plugin ecosystem, see awesome-dsh-plugin.

---

👥 Team

Built by the OfferKuai team — an AI job application assistant on a mission that "users need results, not repeated conversations". Founder: Zhaofeng (Yaming). This plugin is released as open source for the DeepSeek Harness community.

📄 License

MIT