dsh-prompt-enhancer: A One-Click Prompt Enhancer DSH plugin for DeepSeek Harness
fishsb/dsh-prompt-enhancer
DSH plugin to enhance prompts with one click, directly replacing input, with multiple modes and memory chain.
Turn vague draft prompts into clear, structured versions with one click, supporting multiple optimization modes and a memory chain to make prompt polishing in DeepSeek Harness more efficient.
dsh plugin --profile web add github:fishsb/dsh-prompt-enhancer- Category
- Sessions & Messages
- Platform
- DSH-Plugin
- Author
- fishsb
- Distribution
- Plugin
dsh-prompt-enhancer Key Features
dsh-prompt-enhancer Repository
dsh-prompt-enhancer Repo Summary
dsh-prompt-enhancer is a prompt enhancement plugin (DSH plugin) for DeepSeek Harness, maintained by Fishsb under the MIT license, last updated in 2026-08. It solves the problem of vague draft prompts that require repeated manual polishing: click the ✨ button in the input box, the plugin calls an independent LLM to enhance the draft, directly replaces the input box content when done, and allows one-click undo if unsatisfied. Core capabilities include 5 optimization modes (Basic, Light, Standard, Smart, One-Click Publish), an optional memory chain feature, language following (Chinese/English), and strict privacy guards.
What optimization modes does dsh-prompt-enhancer support?
The plugin offers 5 optimization modes: Basic mode sends directly for fastest response; Light mode uses local rules; Standard mode combines rules with workspace/session retrieval; Smart mode adds LLM task progress analysis and full retrieval; One-Click Publish mode starts from a rough idea, automatically plans retrieval topics, performs multi-step web searches, and generates a complete development specification with a nine-chapter structure, automatic scenario routing (game/software), and self-evaluation of the plan. Mode switching takes effect immediately and persists; One-Click Publish mode has no output limit, with timeout automatically extended to 240 seconds or more.
How are memory chain and privacy protection designed in dsh-prompt-enhancer?
The memory feature has an independent switch: when enabled, multiple rounds of 'optimize → modify → re-optimize' accumulate into a memory chain, keeping the most recent 4 rounds with rolling updates; sending or clearing the input box does not clear it, and undo only pops the last round. When disabled, no reading or writing occurs. For privacy, the plugin only injects recent session messages, relevant workspace file summaries, and related session fragments on demand, subject to context budget limits; sensitive files (e.g., .env, keys, credentials, logs) are hard-filtered and never injected. The memory chain is stored only as a boolean flag in browser localStorage; content exists only in current page memory and is sent to the selected model provider with optimization requests.
How to install dsh-prompt-enhancer?
The recommended one-command installation (prerequisite: DeepSeek Harness installed locally and pnpm in PATH):
dsh plugin --profile web add github:fishsb/dsh-prompt-enhancer#v3.0.0
After installation, restart DSH (dsh web); the ✨ button appears in the input box toolbar, indicating success. You can also download a tgz package from the Releases page for offline installation, or use dynamic Cordis installation (read plugin-host.js and plugin-client.js, define with cordis_define, and run with cordis_run). Update and uninstall commands:
dsh plugin --profile web update dsh-prompt-enhancer
dsh plugin --profile web remove dsh-prompt-enhancer
What configuration options does dsh-prompt-enhancer have?
In Settings → 'Models & Plugins' → 'Optimization Parameters', you can configure: optimization mode, memory feature toggle, context budget (0/2000/4000/8000 characters, 0 means no injection), timeout, token limit, output limit, and templates (built-in or custom). The model chain is configured in the 'Model Configuration' tab, supporting sequential attempts, add/delete/reorder, per-step thinking toggle and level, inline connectivity tests, and restore defaults.
Which DeepSeek Harness versions and platforms are compatible with dsh-prompt-enhancer?
The plugin depends on DSH runtime-injected APIs (llm / slots / harness / inputActions / sessionQuery / fs), which may change with DSH version upgrades. Version detection and one-click updates are done by the browser directly connecting to api.github.com (CORS available, host does not need outbound network); in restricted networks, ensure the browser can access GitHub. The plugin itself does not record or report any data; diagnostic logs contain only metadata like mode and duration.