AI Agent Hub
Back to plugins
dsh-custom-instructions preview

dsh-custom-instructions

Client Updated 2026.08.25

Run the following command in DeepSeek Harness:

dsh plugin install lyke-61/dsh-custom-instructions

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

Run dsh plugin install lyke-61/dsh-custom-instructions in your terminal to install the plugin, sourced from https://github.com/lyke-61/dsh-custom-instructions, then restart dsh web once to activate it.

About this plugin

Opening a new conversation in DeepSeek Harness and retyping the same project context, preferred coding style, and output format for every new session and every model switch — the repetition compounds and slows you down. dsh-custom-instructions brings the Codex AGENTS.md / ChatGPT Custom Instructions idea into Harness: write your notes once in Settings General Custom Instructions, and every future conversation on this machine, across every model, carries them automatically.

What the plugin does is deliberately narrow. A monospace editor reads and writes the global instruction file directly; saving flushes to disk with immediate success or failure feedback. Injection rides on DSH built-in dsh-agent-instructions, so new sessions pick up the text from the very first message, and the current session picks it up after the next file read or write. The whole thing ships as a hand-written client bundle plus a pure Node ESM host — no build step, no extra tooling required.

If you live in DeepSeek Harness for development, writing, or data work and want a small fixed context block that stays on the machine and travels with every conversation, this fills the gap that was missing. If you prefer composing ad-hoc prompts in each message, the panel stays out of the way entirely.

Screenshots

Use Cases

  • Every new conversation automatically inherits your project context and coding style
  • Switching models without re-typing output preferences
  • Maintaining a single global instruction set from the settings panel

Best For

  • Engineers doing daily development in DeepSeek Harness
  • Writers who need consistent style across sessions
  • Data analysts who want uniform behavior across all models