AI Agent Hub
Back to plugins
dsh-suggest-ghost preview

dsh-suggest-ghost

Client Updated 2026.09.04

Run the following command in DeepSeek Harness:

dsh plugin install WuJiaoJue/dsh-suggest-ghost

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

Run dsh plugin install WuJiaoJue/dsh-suggest-ghost in DeepSeek Harness to install, source available at https://github.com/WuJiaoJue/dsh-suggest-ghost; after installation, restart dsh web and hard-refresh the page to get started.

About this plugin

Multi-turn conversations in DeepSeek Harness Web often follow predictable patterns: you just reviewed a file and the next prompt is almost certainly to run tests or write docs; the same command recurs with a tweaked argument. dsh-suggest-ghost turns that repetition into zero keystrokes. It is a pure plugin that hooks into host turn events and renders a DOM overlay on the client, with no core code modified and no configuration required.

Two modes switch automatically. When the input box is empty, after each completed turn the last exchange is anonymized and sent to a helper LLM (inheriting your main request routing by default) to predict your next most likely prompt, surfaced as grey ghost text in the input field, much like the suggestion experience in Claude Code. When you have already typed a few characters, the plugin matches prefixes from the current session history, scores by recency and frequency, and completes the remainder in the spirit of zsh autosuggestions. Press Tab to accept the whole suggestion, or the right arrow to adopt word by word; simply keep typing and the ghost vanishes, so there is zero pressure.

Credentials are stripped before any request leaves the client, outputs are sanitized, and every stage is bounded in bytes, tokens, and timeout. Built for developers who operate DSH Web at high frequency and want fewer repetitive keystrokes, with everything ready out of the box.

Screenshots

Use Cases

  • Repeating similar commands across debugging turns, prefix autocomplete cuts down redundant typing
  • After a turn ends and you are unsure of the next step, LLM prediction suggests the most likely prompt
  • Typed a few characters but forgot the previous phrasing, grey ghost text completes the remainder instantly

Best For

  • Developers running high-frequency multi-turn conversations in DSH Web
  • Engineers looking to reduce repetitive typing and keep conversation flow fast
  • Plugin users who prefer zero-config, out-of-the-box experiences