AI Agent Hub
Back to plugins
⚙️

dsh-review-squad

Workflow Updated 2026.08.30

Run the following command in DeepSeek Harness:

dsh plugin install luomeii/dsh-review-squad

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install luomeii/dsh-review-squad inside DeepSeek Harness to install this plugin; the source repository is available at https://github.com/luomeii/dsh-review-squad .

About this plugin

Most existing dsh git plugins stop at diff viewers and worktree cleanup, but code review is the workflow other agent ecosystems keep voting for with stars. dsh-review-squad fills that gap: a single /review command dispatches four independent read-only reviewer subagents for security, correctness, tests, and style. Each reviewer gathers its own evidence by reading the diff, tracing callers, and checking conventions, then all findings are merged into one severity-grouped report. A failed reviewer is reported individually without dragging the whole run down.

Three core capabilities set it apart. First, reviewers are dsh spawn subagents with fresh context and their own tool sets, so they find evidence independently rather than relying on prompt-fed input, making the review substantially deeper than a question over a diff. Second, per-reviewer model routing lets you put your strongest model on security and a cheap fast model on style, balancing cost and quality per concern. Third, the squad runs as a background job by default so the chat input is never locked, the report is aggregated deterministically with no extra LLM call, and a hard character cap prevents output bloat.

This plugin is for developers already using dsh for daily work who want code review embedded in their workflow without manually scanning every diff. Whether you want the agent to self-review before a commit or you want team roles routed to different models, it works out of the box, and every parameter can be tuned via settings.yaml hot-reload without restarting dsh.

Use Cases

  • Dispatch parallel multi-role reviewers to gather evidence before committing
  • Route different review dimensions to different models to balance cost and quality
  • Run reviewers as background jobs so the chat input stays unblocked and the report is ready on demand

Best For

  • Developers using dsh for daily work who want code review embedded in their workflow
  • Engineering teams that want per-role model routing to control inference cost
  • Engineers who need a unified multi-dimension report covering security, correctness, tests, and style