liangmianya/dsh-synapse
liangmianya/dsh-synapseDeepSeek Harness 的可视化非线性对话工作区:把会话、追问与分支变成可浏览、可拖拽的对话地图。
Git 式会话分支,模拟真实程序员协作的并行开发:根分支充当秘书维持调度,子分支持有各自上下文并行工作,任务完成后仅以压缩摘要 squash/rebase 合并回来;提供 /branch 命令、分支页签、send_message_by_branch 跨分支消息,并附一份把会话分支绑定到 git worktree 的治理基线。
English | 简体中文
Sub agents and parallel development never quite deliver in today's agent apps. This project manages dsh's native, discrete sessions the way git manages branches: fork gives conversations inheritance, while squash and rebase give them merging — a major upgrade to the parallel-development and prompt-management experience on dsh.
This is a plugin for DeepSeek Harness; it cannot run standalone.
Version 0.4.0 supports dsh 0.2.1-rc.2. It replaces the previous dsh 0.1.2-rc.1 compatibility line; upgrade dsh before upgrading this plugin.
For compatibility development, build a sibling ../deepseek-harness checkout and run bun run test:harness. Set DSH_HARNESS_DIR to use another checkout. The check uses an isolated copy linked to the actual harness packages and checks V4 messages, historical merges, and both Host and Client types.

main) stays your secretary and dispatcher. Research and coding tasks are delegated to child branches; when a task is done, only its compressed context is reported back to the main branch.fork and cross-session messaging experience.Install (requires a web-app-based dsh profile):
dsh plugin --profile web add dsh-session-fork
After that, just let your agent use the plugin freely — every command also ships as an agent-callable tool!
branch operations give every session a name, an ancestry, and an index — manageable through commands.fork hardens the native experience and provides the ancestry primitive.squash and rebase offer two forms of cross-branch merging.send_message_by_branch strengthens communication between sessions.What we want to build next:
Issues — there are plenty of long-term improvements and bugs waiting.We take an open stance on AI collaboration: feel free to use AI to contribute code, write commit messages, and draft PRs. But we expect you to own your code — review it yourself, and treat AI as your tool in communication rather than letting it talk to us on your behalf.