DSH Side Chat: Open a Parallel Side Conversation alongside the Main Dialogue
DSH Side Chat is an MIT-licensed plugin designed for DSH, aiming to address the issues of overly long and cluttered main conversation contexts. The plugin provides parallel side conversations with independent context windows, enabling context isolation and preventing temporary tasks from polluting the main conversation. Core features include truly parallel agent sessions, side panel management (supporting multi-session switching), Markdown rendering, panel width adjustment (300-900px), and minimized background operation. Users can use the "bring back to main conversation" feature to have the side agent automatically summarize and inject a collapsed summary into the main session, keeping the main conversation flow clean. The plugin supports two installation methods: a dynamic Cordis plugin mode suitable for debugging and an npm package mode suitable for daily use. Typical usage involves creating side conversations via the panel or the `/side` command, with each main session
Read MoreDSH-BTW: DSH Bypass Q&A Plugin
`dsh-btw` is a side-channel Q&A plugin for DSH, designed to address the need for ad-hoc questions during the main task without disturbing the main model's context. Through the `/btw` command and a centered pop-up panel, it provides an independent session branch for read-only Q&A. Core features include: incremental synchronization of the main conversation history for multi-turn dialogue, clearing branches, and independent maintenance per session; branches respond in read-only mode, with the tool allowlist restricted to read, glob, grep, read_image, and web_search, rejecting operations such as file writing and command execution; branches do not enter the main LLM context and are not displayed as sub-agents. Installation requires the dsh CLI and pnpm to be installed locally, followed by executing specific commands to add the plugin and restarting the `dsh web` process. The applicable scenario is for DSH Web users to retain read-only retrieval capabilities while avoiding pollution of the main context.
Read More