Tool plugins for DeepSeek Harness
Give the agent new things it can do.
Tool plugins add actions the agent can call on its own: reading and writing files, running commands, fetching web pages, querying a database, driving a browser. Each one widens what the agent can do without you copying results back and forth. Each one also widens what can go wrong, so this is the category to install with the most care.
5 plugins
- archifytt-a1i Describe a system or point your agent at a repository, and archify returns a self-contained HTML diagram you can explore, refine in chat and export as an image. Tools 95.6k
- Reactive Resumereactive-resume Write and customize resumes and cover letters, review them against job postings, and organize applications. Supports document exports and self-hosting. Tools 45.7k
- next-ai-draw-ioDayuanJiang Generate editable draw.io diagrams from text, images or documents. Use chat and canvas editing together, with version history and an MCP server for AI agents. Tools 36.9k
- huashu-designalchaincyf A design skill for Claude Code, Cursor, Codex and DSH. It sketches six directions, builds the three strongest, and delivers HTML, MP4, GIF, editable PPTX or PDF. Tools 29.2k
- dsh-opsT-Auto A low-level tool plugin for DeepSeek Harness. It adds an ops tool family for Native and PTC modes, loads rarely used tools on demand, and can stand in for PowerShell with Bash on Windows. Tools 5.7k
How to choose
Read what the tool can reach before anything else: which directories, which network hosts, which credentials. A narrow tool that does one job is easier to trust than a broad one that can run anything. Check whether calls need your approval or run automatically, and keep in mind that every extra tool description takes up context and gives the model more ways to choose wrongly.
Terms used on this page
- Tool call
- A structured request from the model to run a named function with specific arguments.
- Permission scope
- The set of files, commands or network targets a tool is allowed to touch.
- Sandbox
- An isolated environment that limits what a tool can affect on the host machine.
Questions and answers
Is it safe to let the agent run shell commands or edit files?
It is as safe as the limits around it. Use approval prompts or a sandbox where the plugin offers them, work in a directory under version control, and avoid granting access to anything you could not afford to lose.
Does installing more tools make the agent more capable?
Up to a point. Past that, the model spends context reading tool descriptions and picks the wrong tool more often. Install what the current work needs and remove what you have stopped using.