dsh-handoff

Hand your long session over to a fresh one: /handoff summarizes the recent conversation with an LLM, creates a clean new session in the same workspace, injects the summary as its first message, and auto-opens it. The origin session is unchanged.

AI 分析

核心用途是解决长对话导致的上下文臃肿和 Token 浪费问题。适合会话轮数过多、需要“瘦身”并继续在当前工作区开展任务的用户。通过总结无缝承接上下文。

パッケージ

dsh-handoff

バージョン

0.1.0

ライセンス

MIT

最終更新

2026/08/16

Usage

Type in any conversation's input box:

/handoff          # summarize the last 8 user messages (default)
/handoff 20       # summarize the last 20 (same as turns=20)
/handoff all      # summarize every user message (same as turns=all)

What happens:

A command card shows "summarizing…";
An LLM produces the handoff document (goal / progress / decisions / files / next steps / open questions), which becomes the new session's first message;
The new session auto-opens; the origin session stays intact and is linked back.

⚠️ Privacy: the recent conversation text (user + assistant messages in the requested window) is sent to the configured model to produce the summary.

Configuration

- id: handoff
  name: dsh-handoff
  config:
    turns: 8            # user-message rounds to extract (1–50)
    maxInputChars: 24000  # byte cap for the summary input (oldest dropped first)
    maxTokens: 2000     # summary output budget
    reasoningEffort: off  # off (default, fast/cheap) | high | max | inherit
    timeoutMs: 120000   # summary call timeout
    # provider: deepseek-official   # optional fixed model route
    # model: deepseek-v4-flash      # must be paired with provider

Without provider/model, the current session's latest logged model route (session.requestHeader()) is used.

dsh-handoff

AI 分析

インストール

ドキュメント

Usage

Configuration