dsh-default-prompt
Run the following command in DeepSeek Harness:
dsh plugin install new-Beginner/dsh-default-prompt
Paste the following prompt into your AI chat to install this plugin:
Run dsh plugin install new-Beginner/dsh-default-prompt in the terminal, or download the Release package from https://github.com/new-Beginner/dsh-default-prompt for offline installation, then restart DSH to activate.
About this plugin
Under the default mechanism of DeepSeek Harness, user-defined global rules or AGENTS.md files are wrapped as a User Role Message and injected into the conversation history, with an explicit clause stating they must not override system instructions. In long multi-turn sessions, models tend to de-prioritize or even ignore these global constraints. If you rely on custom rules to shape model behaviour, this hidden demotion in injection hierarchy is an often-overlooked pain point.
dsh-default-prompt follows the OpenAI Codex paradigm by registering core instructions directly into the underlying System Prompt via systemPrompt.section, placed immediately after the persona definition to achieve the highest system-level authority. The plugin supports flexible combinations of a global file, project workspace overrides, and inline editing, and automatically strips the section when content is empty so no extra tokens are consumed. A built-in Web console provides real-time preview of the final injected text with word counting, and a one-click import can migrate local Codex AGENTS.md rules without manual copy-paste.
This plugin is best suited for developers who heavily rely on global custom rules within DeepSeek Harness, especially those migrating from the OpenAI Codex ecosystem who want to preserve the Codex-style System Prompt workflow. Existing prompt files keep their structure; after installation and a simple restart, rules gain genuine system-level enforcement.
Use Cases
- Global rules being de-prioritized or ignored in long multi-turn conversations
- Preserving the existing rules system after migrating from OpenAI Codex to DeepSeek Harness
- Achieving the highest system-level authority for core behavioural constraints
Best For
- DSH developers who heavily rely on global custom rules
- Engineers migrating from the Codex ecosystem who want to preserve their workflow
- Teams that need precise control over model instruction behaviour
Related Plugins
A unified suite combining hot runtime injection, task-aware thinking-mode routing, and a graded session protocol with red-team gates to sustain model diligence across long-horizon inference.
ModLens is a vision plugin for DeepSeek Harness that gives text-only models sight by reading images pasted directly into chat, with zero-config setup and multiple vision engines.
On-demand vision for text-only DeepSeek Harness agents: built-in free keyless vision chain and 14 vision tools, routing image turns as tool calls to vision models with pixel fidelity, no Python needed, one-command install.
Give text-only models in DeepSeek Harness eyes, enabling image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks.