dsh-context-branch
Run the following command in DeepSeek Harness:
dsh plugin install yummy4727/dsh-context-branch
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install yummy4727/dsh-context-branch in your terminal; the source is at https://github.com/yummy4727/dsh-context-branch. Restart the running DeepSeek Harness web profile after installation to activate the plugin.
About this plugin
In a DeepSeek Harness engineering session, the conversation is linear: after dozens or even hundreds of tool calls, you suddenly want to clarify a term in an earlier answer. The two options are both unsatisfying—ask in the main thread and an unrelated question gets re-sent with every subsequent request, or open a new conversation and lose all the context that made the question meaningful. dsh-context-branch offers a third path: fork from the specific answer you want to build on, and the follow-up becomes an isolated branch the main conversation never sees.
Each branch is a real, independent session that inherits every message up to the fork point and nothing after. It runs read-only: it can inspect workspace files but cannot write or execute mutations, so a follow-up can never disturb the engineering work still in progress in the main thread. The request prefix is byte-for-byte identical to the main conversation at the moment of the fork, so the provider KV-cache still hits and the inherited history is not re-billed. The tree details panel preserves every intermediate tool call, tool result, and reasoning block—click any node to replay the full path.
If you run multi-turn engineering tasks in DeepSeek Harness, need to keep asking clarifying questions deep into a long conversation, and care about input cost (linear accumulation grows as O(N²) while branch mode stays near O(N)), this plugin is a ready-to-use solution. It leaves the standard chat view, sidebar, and session list untouched; branches never appear as separate sessions, and switching back to Chat drops you straight into the main thread where you left off.
Use Cases
- Ask a clarifying question deep in a long engineering conversation without polluting the main thread
- Compare multiple solutions to the same problem in isolated branches that cannot see each other
- Replay the full tool calls, results, and reasoning blocks of any step for audit and reproduction
Best For
- Power users running multi-turn engineering sessions in DeepSeek Harness
- Developers who care about token cost and want to leverage KV-cache hits
- AI-workflow builders who need repeated follow-ups, branch exploration, and full step-level audit trails
Related Plugins
A method pack that makes AI coding agents plan against your real baseline, prove completion with fresh evidence, and reduce reworks and unsafe changes.
Turns the DeepSeek Harness session into a captain that builds a durable sub-agent team, splits goals into dependency-aware tasks, and coordinates work via direct messages and a live Web UI.
Gives coding agents design judgment, letting Claude Code, Cursor, and 70+ agents generate and iterate high-quality UI, presentations, and graphics right from the terminal.
Run the Pi ecosystem's plugins on DeepSeek Harness, unmodified, via a compatibility layer that implements Pi's public extension ABI on DSH's native services.