morlay/deepseek-harness — explained in plain English
Analysis updated 2026-05-18
Improve DeepSeek model performance in OpenCode by using Chinese-localized tool descriptions.
Apply bulk code changes with a custom patch tool that fixes a gap in OpenCode's default tooling.
Search code structurally using the bundled ast-grep based tool.
Debug failed tool calls or inspect raw API traffic using the built-in logging and proxy mode.
| morlay/deepseek-harness | adrianhajdin/job_pilot | cneuralnetwork/pracdago | |
|---|---|---|---|
| Stars | 20 | 20 | 20 |
| Language | TypeScript | TypeScript | TypeScript |
| Setup difficulty | easy | hard | moderate |
| Complexity | 3/5 | 4/5 | 2/5 |
| Audience | developer | developer | vibe coder |
Figures from each repo's GitHub metadata at analysis time.
Requires an existing OpenCode installation configured to use a DeepSeek model.
DeepSeek Harness is a plugin for OpenCode (an AI-powered coding assistant) that is specifically tuned for DeepSeek language models used in Chinese-language environments. The core idea is that DeepSeek performs better when it receives prompts and tool descriptions in Chinese rather than English, so this plugin translates all of OpenCode's built-in tool descriptions into Chinese and removes redundant parameter explanations to let the model reason and respond more naturally. Beyond language, the plugin also patches gaps in how OpenCode's built-in tools work with DeepSeek, for example, fixing a "patch" tool that did not apply correctly. It ships 11 localized built-in tools, two custom tools (one for applying bulk code changes via unified diff format, another for structured code search using ast-grep), and four agent roles (a main code agent, a chat agent, a documentation agent, and a sub-agent worker for delegated tasks). It also logs every tool call that fails to a JSON lines file for debugging, and includes an optional transparent proxy mode that captures raw API requests and responses for developers who want to inspect exactly what the model is sending and receiving. The plugin is written in TypeScript and installed by adding a single line to an OpenCode configuration file.
An OpenCode plugin that translates tool descriptions into Chinese and patches gaps so DeepSeek AI models work better in Chinese-language coding.
Mainly TypeScript. The stack also includes TypeScript, OpenCode, DeepSeek.
Setup difficulty is rated easy, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.