{"id":558933,"date":"2026-10-29T07:59:13","date_gmt":"2026-10-29T07:59:13","guid":{"rendered":"https:\/\/webkul.com\/blog\/?p=558933"},"modified":"2026-10-29T07:59:18","modified_gmt":"2026-10-29T07:59:18","slug":"deepseek-harness-agent-framework","status":"publish","type":"post","link":"https:\/\/webkul.com\/blog\/deepseek-harness-agent-framework\/","title":{"rendered":"DeepSeek Harness: Everything Really Is a Plugin"},"content":{"rendered":"\n

DeepSeek Harness<\/strong> just open-sourced a different way to build AI <\/a>agents. Instead of one fixed coding assistant, DeepSeek released the scaffolding underneath it.<\/p>\n\n\n\n

Every piece is swappable. The model, tools, sandbox, session store, and agent loop all exist as plugins.<\/p>\n\n\n\n

These plugins mount into a shared runtime called Cordis. As a result, developers can reconfigure an agent without touching its source code.<\/p>\n\n\n\n

\"DeepSeek<\/figure>\n\n\n\n

Image Source: DeepSeek Harness plugin<\/a><\/em><\/p>\n\n\n\n

The Challenge with Fixed Agent Frameworks<\/h2>\n\n\n\n

Most coding agents bundle tools, memory, planning, and sub-agent logic into one rigid shell. Consequently, changing one piece often breaks another.<\/p>\n\n\n\n

For instance, swapping the model provider means editing core logic. Changing the sandboxing strategy causes the same problem.<\/p>\n\n\n\n

Overall, this tight coupling makes agents hard to extend. It also makes safe experimentation difficult.<\/p>\n\n\n\n

DeepSeek Harness takes the opposite approach. Its kernel, Cordis, contains no agent capabilities at all.<\/p>\n\n\n\n

Instead, Cordis only handles plugin registration, dependency resolution, and lifecycle management. Every real capability lives in a plugin, from file editing to the UI.<\/p>\n\n\n\n

Therefore, replacing the model backend doesn’t touch the planning plugin. Likewise, swapping the tool plugin doesn’t affect the UI.<\/p>\n\n\n\n

Four Runtime Modes for Four Different Jobs<\/h2>\n\n\n\n

Rather than forcing one setup on every use case, DeepSeek Harness ships four preset runtime modes. Switching between them is a configuration choice, not a reinstall.<\/p>\n\n\n\n

Mode<\/th>What it loads<\/th>Best for<\/th><\/tr><\/thead>
Standard<\/strong><\/td>Full toolset \u2014 file editing, shell access, web search, skills, planning, goals, subagents, and workflows<\/td>General-purpose coding agents<\/td><\/tr>
Code (PTC)<\/strong><\/td>Standard mode, exposed through a TypeScript SDK so the model writes one program instead of many tool calls<\/td>Complex, multi-step tool workflows<\/td><\/tr>
Minimal<\/strong><\/td>Just a shell tool and a file editor<\/td>Benchmarking models in a lean environment<\/td><\/tr>
Creator<\/strong><\/td>Standard mode plus runtime inspection, in-memory plugin experiments, and preset authoring<\/td>Building and testing new plugin combinations<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

Code mode solves a real, recurring problem. Normally, checking fifty files for a TODO comment takes fifty round-trips.<\/p>\n\n\n\n

In Code mode, however, the model writes one short TypeScript program. That program loops through the files itself.<\/p>\n\n\n\n

Only the final result goes back to the model. Consequently, fifty round-trips collapse into one execution.<\/p>\n\n\n\n

Creator mode, meanwhile, turns the harness into its own development environment. It exposes runtime inspection and live plugin authoring.<\/p>\n\n\n\n

As a result, a developer can ask the agent to build a new plugin. For instance, it could create a small game window in the UI corner.<\/p>\n\n\n\n

The agent then inspects the environment and writes the code. Finally, it packages the plugin and mounts it into the running session.<\/p>\n\n\n\n

Full Traceability Through the Trajectory View in DeepSeek Harness<\/h2>\n\n\n\n

Debugging an agent from raw console output is painful. To address this, DeepSeek Harness uses an append-only session log.<\/p>\n\n\n\n

A simple runtime rule enforces it: whatever the model sees gets logged. For example, that includes system prompts, chain-of-thought, tool calls, and context injections.<\/p>\n\n\n\n

This log powers the Trajectory view. It lets developers inspect exactly how a run unfolded, step by step.<\/p>\n\n\n\n

Beyond viewing, trajectories can be:<\/p>\n\n\n\n