dsh-fork-diff
Run the following command in DeepSeek Harness:
dsh plugin install chouyong/dsh-fork-diff
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install chouyong/dsh-fork-diff in your DeepSeek Harness terminal to complete installation; source code is available at https://github.com/chouyong/dsh-fork-diff
About this plugin
When you fork a session in DeepSeek Harness, the real question is rarely about which branches exist; it is about what exactly differs between the two public histories. Opening each session and scrolling back and forth makes it hard to compare long answers, tool-call trajectories, and token costs at a glance. dsh-fork-diff collapses all of that into a single filterable two-pane view.
Under the hood, the plugin auto-discovers parent, child, and sibling branches, then aligns user messages, final assistant answers, tool calls, and tool results segment by segment. Line-level highlighting distinguishes unchanged, modified, left-only, and right-only content, while summaries cover elapsed time and visible token usage. Large histories explicitly enter a fast-alignment mode rather than claiming approximate results are exact. Everything is strictly read-only: no forks are created, no session events are written, and no conversation content is uploaded.
Built for Agent experimenters who frequently fork branches to A/B compare runs: after a multi-path conversation, click the Compare Branch button, pick a counterpart, switch to Diff-only mode, and pinpoint the divergence in seconds before deciding which path to keep.
Screenshots
Use Cases
- Compare answer and tool-call differences across forked branches
- Identify which branch deviated from expected behavior in multi-path runs
- Verify whether a child branch inherited key context from its parent
Best For
- Agent experimenters who fork branches for A/B comparison
- Developers auditing tool-call trajectories and token overhead
- Researchers managing multi-session branch workflows
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.

