dsh--prompt--enhance: A Prompt Enhancement Plugin for DeepSeek Harness
sunzhentao/dsh--prompt--enhance
A DSH plugin that enhances prompts with basic, standard, and expert modes, rewriting prompts with project context, and offering comparison confirmation and undo.
dsh--prompt--enhance is a prompt enhancement plugin for DeepSeek Harness that loads automatically with the DSH Web profile, no manual reload needed. It offers three enhancement modes—Basic, Standard, and Expert—for pure text polish, context-aware rewriting, and deep optimization, while automatically adapting reasoning effort to different models and gateways to avoid unstable output. Enhanced results are confirmed via a compare panel with 9-second undo, and pending confirmation items are shown separately, ensuring the clean text is always what gets sent to the LLM.
How to Install
dsh plugin --profile web add @lidaxi/prompt-enhance- Category
- Tools & Capabilities
- Platform
- DSH Plugin
- Author
- sunzhentao
- Distribution
- Plugin
dsh--prompt--enhance Key Features
dsh--prompt--enhance Repo Summary
What Does It Do?
prompt-enhance is a DSH plugin for DeepSeek Harness that enhances prompts directly inside the harness. It solves the problem of inconsistent prompt quality and unstable output caused by model differences, letting you polish, rewrite with project context, or deeply optimize your input without manual reloading. Core capabilities include three enhancement modes, automatic adaptation to different models/gateways, and a confirmation loop for unresolved items.
Core Features
- Three enhancement modes: Basic (pure text polish, no project context), Standard (rewrites with directory structure, dependency config, and entry files), Expert (deeper context + task analysis + verifiable acceptance criteria).
- Universal model adaptation: no vendor hardcoding; runtime detection and graceful degradation of reasoning effort parameters (e.g.,
reasoningEffort), with automatic retry or fallback to a candidate model chain (up to 3). - Compare and confirm: after enhancement, a "Original | Enhanced" panel appears; only clicking "Adopt" fills the input, and you can undo within 9 seconds.
- Pending confirmation loop: "to-confirm" items are stripped from the body and shown separately; you can answer each and re-enhance, and the clean text is always what gets adopted/sent.
- Busy status card with cancel: shows "Enhancing (mode)… elapsed N s" with a cancel button that truly interrupts the server-side stream.
- Configurable: mode parameters, candidate model chain, etc., can be overridden via the composition row config, requiring zero code changes for deployment.
How to Use This Plugin?
After enabling it in DSH, type a prompt in the input box and click the ✦ button to trigger enhancement. No configuration is needed by default; to adapt to specific models/gateways, write the full config in $DSH_HOME/cordis.patch.yml or a --patch overlay (the config for the same row id is replaced entirely), e.g., set maxAttempts, candidates, and per-mode effort, maxTokens, temperature. After enhancement, choose "Adopt" or "Discard" in the compare panel, or click "↩ Undo" within 9 seconds to restore the original.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: sunzhentao/dsh--prompt--enhance. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.