@jianyiagi/dsh-prompt-enhancer 0.0.0-stage → 1.0.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 jianyiagi
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -1,3 +1,111 @@
-# Temporary Holding Version
+# @jianyiagi/dsh-prompt-enhancer
 
-This version is a temporary placeholder for this package. An operational version to replace this has been submitted for review and is awaiting a staged release.
+> DSH（DeepSeek Harness）Web 端插件：**输入框一键优化提示词**。
+> 有内容时，模型选择器左侧出现一个闪光按钮；点一下，模型把提示词改写得更清晰、更可执行，结果直接替换输入框草稿。
+
+---
+
+## 功能
+
+- **一键增强**：优化表达 + 结构化（自然段落优先），消除歧义与口语化，明确目标 / 对象 / 约束 / 期望输出
+- **忠于原文**：保留原意与语言，不编造原文没有的信息；代码、命令、路径、变量名、字段名**原样保留**
+- **无冗余**：不写「未指定 / 未提供」占位，各部分不重复，长度不明显超过原文
+- **直接可用**：结果写回输入框，不满意可直接编辑器撤销
+- **双语**：界面文案随 DSH 语言（中文 / English）
+
+---
+
+## 安装
+
+### 桌面端（Electron App）
+
+**推荐**：左侧栏「**插件**」页 → 安装 bundle → 填入包名：
+
+```
+@jianyiagi/dsh-prompt-enhancer
+```
+
+装完**完全退出 App 再重新启动**（不是刷新页面）。
+
+### CLI / Web 端
+
+```powershell
+dsh plugin --profile <你的profile> add @jianyiagi/dsh-prompt-enhancer
+```
+
+若 `dsh` 不在 PATH，用 DSH 安装目录下的 bin：
+
+```powershell
+node "<dsh-app>\lib\bin.js" plugin --profile=<profile> add @jianyiagi/dsh-prompt-enhancer
+```
+
+> `<dsh-app>` = DSH 安装目录下的 `resources\dsh\app`。
+> 若装完没生效，说明包已安装但 **bundle 未被选中**：在「插件」页勾选它，或把包名加入 profile 的 `dsh.profile.bundles` 后重启。
+
+---
+
+## 使用
+
+| 入口 | 行为 |
+|---|---|
+| 输入框工具行的闪光按钮（模型选择器左侧） | 草稿非空时出现 → 单击 → 转圈 → 草稿被增强结果替换 |
+| 斜杠命令 `/prompt-enhance <文本>` | 结果以命令结果行展示，**不写回草稿**（附带能力） |
+
+增强失败时按钮左侧显示一行错误（4 秒后自动消失）。
+
+---
+
+## 配置：固定增强所用模型（推荐）
+
+默认**跟随会话当前模型**。若想更快，可在 profile 补丁里固定一个轻量模型：
+
+```yaml
+- insert:
+    - id: prompt-enhancer
+      name: '@jianyiagi/dsh-prompt-enhancer'
+      config:
+        provider: <provider-id>
+        model: <model-id>
+```
+
+两个字段必须同时提供才生效；删掉 `config` 即恢复跟随会话模型。
+
+> 为什么建议固定：增强是**生成密集型**调用。实测同一段提示词，大模型需 25～48 秒，轻量模型仅 4～6 秒（约 6～12 倍）。输出很短，瓶颈完全在模型思考。
+
+---
+
+## 已知限制（不是 bug）
+
+1. **每次增强会在对话流留一行** `prompt-enhance · <结果预览>`。原因：插件唯一的客户端→宿主通道是命令注册表，注册表强制记录 `command/run`/`command/done`，而 Chat 层的命令行展示是宿主内建行为、未开插件级隐藏接口。该行不影响模型历史、不计 token。
+2. **引用芯片会退化为纯文本**：草稿里的 `@文件` 引用在整体替换后变成普通文本。
+3. **直接替换、无预览**：不满意请用编辑器撤销。
+4. **仅文本增强**：不改动附件；单次输入上限 24 KB，超时 60 秒。
+
+---
+
+## 卸载
+
+桌面端：「插件」页卸载。CLI：
+
+```powershell
+dsh plugin --profile <profile> remove @jianyiagi/dsh-prompt-enhancer
+```
+
+---
+
+## 它怎么工作
+
+```
+输入框草稿 → 按钮（会话作用域槽位 conversation.input.right）
+   → ctx.remote.commands.execute("/prompt-enhance <草稿>")
+   → 宿主命令用 ctx.llm.stream 调模型（凭据走你的账户登录态，插件不接触 API Key）
+   → 结果经命令结果回传 → inputActions.setDraft(结果)
+```
+
+- 不读取、不存储、不上传任何 API Key
+- 不发送除提示词本身以外的任何内容给模型
+- 开源免费（MIT）
+
+---
+
+**版本**：1.0.0 ｜ **维护者**：@jianyiagi

package/client.js

@@ -0,0 +1,141 @@
+/**
+ * dsh-prompt-enhancer — browser half.
+ *
+ * Injects a one-click enhancement button into the composer tool row
+ * (`conversation.input.right`, immediately left of the model selector).
+ * The button renders only while the draft is non-empty; a click runs the
+ * host `/prompt-enhance` command and replaces the whole draft with the
+ * returned text through the standard `inputActions.setDraft` face.
+ */
+window.__ModuleLoader__.load({
+  id: '@local/dsh-prompt-enhancer',
+  factory(require) {
+    const React = require('react');
+    const h = React.createElement;
+    const NS = 'prompt-enhancer';
+
+    const CSS = [
+      '.pe-error{font-size:12px;line-height:20px;white-space:nowrap;',
+      'color:var(--dsw-alias-state-error-primary);margin-right:6px;',
+      'max-width:240px;overflow:hidden;text-overflow:ellipsis}',
+      '.pe-button{box-sizing:border-box;display:inline-flex;flex:none;',
+      'align-items:center;justify-content:center;width:28px;height:28px;',
+      'padding:0;border:1px solid var(--dsw-alias-border-l2);border-radius:8px;',
+      'background:transparent;color:var(--dsw-alias-label-secondary);cursor:pointer}',
+      '.pe-button:hover:not(:disabled){background:var(--dsw-alias-interactive-bg-hover);',
+      'color:var(--dsw-alias-label-primary)}',
+      '.pe-button:disabled{cursor:default;opacity:.7}',
+      '.pe-button:focus-visible{outline:var(--dsw-focus-ring-width,2px) solid',
+      'var(--dsw-focus-ring-color,var(--dsw-alias-state-business-primary));outline-offset:-1px}',
+      '.pe-spinner{box-sizing:border-box;width:12px;height:12px;border-radius:50%;',
+      'border:2px solid currentColor;border-top-color:transparent;',
+      'animation:pe-spin .6s linear infinite}',
+      '@keyframes pe-spin{to{transform:rotate(360deg)}}',
+      '@media (prefers-reduced-motion:reduce){.pe-spinner{animation:none}}',
+    ].join('');
+
+    const ICON = h('svg', {
+      viewBox: '0 0 16 16', width: 14, height: 14, 'aria-hidden': true,
+      style: { display: 'block' },
+    }, h('path', {
+      fill: 'currentColor',
+      d: 'M8 1.5l1.6 4.4L14 7.5l-4.4 1.6L8 13.5 6.4 9.1 2 7.5l4.4-1.6z',
+    }), h('path', {
+      fill: 'currentColor', opacity: 0.7,
+      d: 'M12.8 10.2l.6 1.6 1.6.6-1.6.6-.6 1.6-.6-1.6-1.6-.6 1.6-.6z',
+    }));
+
+    const SPINNER = h('span', { className: 'pe-spinner', 'aria-hidden': true });
+
+    const ZH = {
+      'button.title': '增强提示词',
+      'error.unknown': '命令未匹配',
+      'error.empty': '模型未返回增强结果',
+      'error.failed': '增强失败',
+    };
+
+    const EN = {
+      'button.title': 'Enhance prompt',
+      'error.unknown': 'Command not matched',
+      'error.empty': 'Model returned nothing',
+      'error.failed': 'Enhancement failed',
+    };
+
+    function EnhanceButton({ useInput, inputActions, enhance, t }) {
+      const draft = useInput((state) => state.draft) ?? '';
+      const [busy, setBusy] = React.useState(false);
+      const [error, setError] = React.useState('');
+      const errorTimer = React.useRef(0);
+      React.useEffect(() => () => window.clearTimeout(errorTimer.current), []);
+      if (draft.trim() === '') return null;
+      const flashError = (message) => {
+        setError(message);
+        window.clearTimeout(errorTimer.current);
+        errorTimer.current = window.setTimeout(() => setError(''), 4000);
+      };
+      const onClick = async () => {
+        if (busy) return;
+        setBusy(true);
+        setError('');
+        try {
+          const result = await enhance(draft);
+          if (!result.ok) {
+            flashError(result.error?.message ?? t('error.failed'));
+            return;
+          }
+          const value = result.value;
+          if (value === undefined) {
+            flashError(t('error.unknown'));
+            return;
+          }
+          if (value.result.kind === 'error') {
+            flashError(value.result.text);
+            return;
+          }
+          const text = (value.result.text ?? '').trim();
+          if (text === '') {
+            flashError(t('error.empty'));
+            return;
+          }
+          inputActions.setDraft(text);
+        } catch (thrown) {
+          flashError(thrown?.message ?? String(thrown));
+        } finally {
+          setBusy(false);
+        }
+      };
+      return h(React.Fragment, null,
+        h('style', { key: 'pe-style' }, CSS),
+        error === '' ? null : h('span', {
+          key: 'pe-error', className: 'pe-error', role: 'status',
+        }, error),
+        h('button', {
+          key: 'pe-button',
+          type: 'button',
+          className: 'pe-button',
+          title: t('button.title'),
+          'aria-label': t('button.title'),
+          disabled: busy,
+          onMouseDown: (event) => event.preventDefault(),
+          onClick,
+        }, busy ? SPINNER : ICON),
+      );
+    }
+
+    return {
+      inject: ['slots', 'locale', 'remote', 'remote.commands'],
+      apply(ctx) {
+        ctx.effect(() => ctx.locale.register(NS, { zh: ZH, en: EN }), 'prompt-enhancer: locale');
+        ctx.slots.inject('conversation.input.right', () => ctx.slots.register({
+          name: 'conversation.input.right',
+          id: 'prompt-enhancer-button',
+          order: 10,
+          locale: NS,
+          inject: (sessionId) => ({
+            enhance: (draft) => ctx.remote.commands.execute(sessionId, `/prompt-enhance ${draft}`, []),
+          }),
+        }, EnhanceButton));
+      },
+    };
+  },
+});

package/cordis.patch.yml

@@ -0,0 +1,8 @@
+# @jianyiagi/dsh-prompt-enhancer profile patch: inserts the plugin row.
+# The host half (index.js) registers the /prompt-enhance command that runs the
+# model-backed enhancement; the browser half (client.js) renders the one-click
+# button in the composer tool row, immediately left of the model selector.
+
+- insert:
+    - id: prompt-enhancer
+      name: '@jianyiagi/dsh-prompt-enhancer'

package/enhance-core.js

@@ -0,0 +1,102 @@
+/**
+ * dsh-prompt-enhancer — pure enhancement core.
+ *
+ * No host imports: everything here is unit-testable in plain Node. The host
+ * half (index.js) wires these helpers to the command handler and the LLM
+ * stream.
+ */
+
+/** Built-in enhancement instruction (not configurable). */
+export const SYSTEM_PROMPT = [
+  '你是提示词工程专家。请对用户给出的提示词做优化与结构化改写，让它更清晰、更完整、更容易被执行。',
+  '',
+  '改写原则：',
+  '1. 优化表达：消除歧义和口语化说法，明确任务目标、对象、约束与期望输出。',
+  '2. 忠于原文：保留原有意图、语言和信息量，不添加原文没有的事实（中文进中文出，英文进英文出）。',
+  '3. 保持人类可读：一律用自然流畅的段落或短句表达，禁止使用小标题分节；读起来要像人写的需求说明。',
+  '4. 简洁优先：改写结果不要明显长于原文；各部分之间不要重复表述，能合并的合并。',
+  '5. 不写占位：原文没有的信息直接省略，禁止输出「未指定」「未提供」「未明确」之类的占位文字。',
+  '6. 原样保留：涉及代码、命令、路径、变量名、字段名等内容时原样保留，不要改写或翻译。',
+  '',
+  '输出要求：只输出改写后的提示词本体，不要解释、前言、后缀，不要使用 Markdown 代码围栏或其他包装。',
+].join('\n');
+
+/** Largest draft the command accepts (bytes). */
+export const MAX_INPUT_BYTES = 24576;
+/** Largest enhancement result the command accepts (tokens). */
+export const MAX_OUTPUT_TOKENS = 4096;
+/** Auxiliary-call deadline (ms). */
+export const TIMEOUT_MS = 60000;
+/** Message-source kind recorded for enhancement requests. */
+export const ENHANCE_SOURCE_KIND = 'prompt-enhancer';
+
+/** Strip the leading separator whitespace `parseCommand` keeps in rawInput. */
+export function trimDraft(rawInput) {
+  return rawInput.replace(/^[\t\n\r ]+/, '');
+}
+
+/** Strip one whole-output Markdown code fence, if the entire output is fenced. */
+export function stripFence(text) {
+  const trimmed = text.trim();
+  const match = /^```[^\n]*\n([\s\S]*?)\n?```$/.exec(trimmed);
+  return match === null ? trimmed : match[1].trim();
+}
+
+/** Normalize raw model output into replacement draft text; `''` when unusable. */
+export function normalizeEnhancedText(raw) {
+  return stripFence(raw).trim();
+}
+
+/** Map a terminal finish reason to an Error, or `undefined` for a clean stop. */
+export function finishError(finish) {
+  switch (finish.kind) {
+    case 'stop': return undefined;
+    case 'error':
+    case 'aborted': {
+      const error = new Error(finish.failure?.message ?? finish.kind);
+      return error;
+    }
+    case 'max-tokens': return new Error('增强结果超出长度上限');
+    case 'tool-calls': return new Error('增强模型意外请求了工具');
+    default: return new Error(`不支持的结束原因 "${String(finish.kind)}"`);
+  }
+}
+
+/**
+ * Read an optional pinned model route from the plugin row's `config`.
+ *
+ * The enhancement call is generation-bound: a large session model can take
+ * tens of seconds, so a deployment may pin a fast model for this auxiliary
+ * call regardless of the session's current model. Both fields must be
+ * non-empty to take effect; otherwise the session route is used.
+ * @returns `{provider, model}` for a complete pin, or `undefined`.
+ */
+export function resolvePinnedRoute(config) {
+  const provider = config?.provider;
+  const model = config?.model;
+  if (typeof provider !== 'string' || provider.trim() === '') return undefined;
+  if (typeof model !== 'string' || model.trim() === '') return undefined;
+  return { provider: provider.trim(), model: model.trim() };
+}
+
+/**
+ * Resolve the model route for the enhancement call, mirroring the Session
+ * Controller's `selectionFor(agent).current` chain.
+ * @returns `{provider, model}`, or `undefined` when nothing resolves.
+ */
+export function resolveRoute(ctx, agent) {
+  const state = ctx.sessionProjections?.stateOf?.(agent.session, 'modelSelection');
+  const pending = state?.pending ?? undefined;
+  if (pending?.provider !== undefined && pending?.model !== undefined) {
+    return { provider: pending.provider, model: pending.model };
+  }
+  const header = agent.session.requestHeader?.();
+  if (header?.config?.provider !== undefined && header?.config?.model !== undefined) {
+    return { provider: header.config.provider, model: header.config.model };
+  }
+  const fallback = ctx.agentDefaultModel?.currentSelection?.();
+  if (fallback?.provider !== undefined && fallback?.model !== undefined) {
+    return { provider: fallback.provider, model: fallback.model };
+  }
+  return undefined;
+}

package/icon.svg

@@ -0,0 +1,4 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 16 16" width="16" height="16">
+  <path fill="#4176E6" d="M8 1.5l1.6 4.4L14 7.5l-4.4 1.6L8 13.5 6.4 9.1 2 7.5l4.4-1.6z"/>
+  <path fill="#4176E6" opacity=".7" d="M12.8 10.2l.6 1.6 1.6.6-1.6.6-.6 1.6-.6-1.6-1.6-.6 1.6-.6z"/>
+</svg>

package/index.js

@@ -0,0 +1,81 @@
+/**
+ * dsh-prompt-enhancer — host half.
+ *
+ * Registers the `/prompt-enhance` command: one auxiliary model call that
+ * optimizes and structures the supplied draft, following the official
+ * session-title-llm auxiliary-call pattern. The browser half calls it
+ * through `ctx.remote.commands.execute`; the command is the only sanctioned
+ * Client-to-Host operation path for a plain-JS bundle.
+ */
+import { createUserMessage, BlockAssembler } from '@deepseek-ai/dsh-llm';
+import {
+  SYSTEM_PROMPT,
+  MAX_INPUT_BYTES,
+  MAX_OUTPUT_TOKENS,
+  TIMEOUT_MS,
+  ENHANCE_SOURCE_KIND,
+  trimDraft,
+  normalizeEnhancedText,
+  finishError,
+  resolveRoute,
+  resolvePinnedRoute,
+} from './enhance-core.js';
+
+export const inject = ['llm', 'commands', 'sessionProjections', 'agentDefaultModel'];
+
+export function apply(ctx, config) {
+  // Optional pinned route from the plugin row's config: the enhancement call is
+  // generation-bound, so a deployment may point it at a fast model instead of
+  // the session's current one.
+  const pinned = resolvePinnedRoute(config);
+  ctx.effect(() => ctx.commands.register({
+    name: 'prompt-enhance',
+    description: '优化并结构化一段提示词文本',
+    input: { hint: '<提示词>' },
+    recordInput: false,
+    async handler({ agent, rawInput, signal }) {
+      const draft = trimDraft(rawInput);
+      if (draft === '') return { kind: 'error', text: '输入为空：没有需要增强的提示词' };
+      const inputBytes = Buffer.byteLength(draft, 'utf8');
+      if (inputBytes > MAX_INPUT_BYTES) {
+        return { kind: 'error', text: `输入过长（${inputBytes} 字节，上限 ${MAX_INPUT_BYTES} 字节）` };
+      }
+      const route = pinned ?? resolveRoute(ctx, agent);
+      if (route === undefined) {
+        return { kind: 'error', text: '无法确定增强所用模型：请先在会话中选择一个模型' };
+      }
+      const messages = [createUserMessage({
+        content: [{ type: 'text', text: draft }],
+        source: { kind: ENHANCE_SOURCE_KIND },
+      })];
+      const assembler = new BlockAssembler();
+      try {
+        for await (const chunk of ctx.llm.stream({
+          provider: route.provider,
+          model: route.model,
+          messages,
+          system: SYSTEM_PROMPT,
+          maxTokens: MAX_OUTPUT_TOKENS,
+          sessionId: agent.session.id,
+          signal: AbortSignal.any([signal, AbortSignal.timeout(TIMEOUT_MS)]),
+        })) {
+          assembler.push(chunk);
+        }
+      } catch (error) {
+        return { kind: 'error', text: `增强失败：${error?.message ?? String(error)}` };
+      }
+      const terminal = finishError(assembler.finish);
+      if (terminal !== undefined) {
+        return { kind: 'error', text: `增强中断：${terminal.message}` };
+      }
+      const text = normalizeEnhancedText(
+        assembler.blocks()
+          .filter((block) => block.type === 'text')
+          .map((block) => block.text)
+          .join(''),
+      );
+      if (text === '') return { kind: 'error', text: '模型未返回增强结果' };
+      return { kind: 'success', text };
+    },
+  }), 'prompt-enhancer: command');
+}

package/locale/en.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "Prompt Enhancer",
+    "description": "One-click prompt optimization and structuring for the composer"
+  }
+}

package/locale/zh.json

@@ -0,0 +1,6 @@
+{
+  "meta": {
+    "title": "提示词增强器",
+    "description": "一键优化并结构化输入框中的提示词"
+  }
+}

package/package.json

@@ -1,6 +1,48 @@
 {
   "name": "@jianyiagi/dsh-prompt-enhancer",
-  "version": "0.0.0-stage",
-  "stub": true,
-  "description": "Temporary package placeholder for staged publishing"
-}
+  "version": "1.0.0",
+  "description": "One-click prompt enhancement for the DSH composer: 输入框一键优化提示词并写回草稿",
+  "type": "module",
+  "icon": "./icon.svg",
+  "exports": {
+    ".": "./index.js",
+    "./client": "./client.js",
+    "./package.json": "./package.json",
+    "./locale/*.json": "./locale/*.json"
+  },
+  "files": [
+    "index.js",
+    "enhance-core.js",
+    "client.js",
+    "cordis.patch.yml",
+    "icon.svg",
+    "locale",
+    "README.md",
+    "LICENSE"
+  ],
+  "peerDependencies": {
+    "@deepseek-ai/dsh-llm": ">=0.2.0-rc.1 <0.4.0"
+  },
+  "dsh": {
+    "bundle": {
+      "patch": "./cordis.patch.yml"
+    },
+    "client": {
+      "platform": "web",
+      "immediately": true,
+      "inject": [
+        "@deepseek-ai/dsh-client-ui-conversation"
+      ]
+    }
+  },
+  "keywords": [
+    "dsh",
+    "deepseek-harness",
+    "plugin",
+    "prompt",
+    "prompt-engineering",
+    "composer"
+  ],
+  "author": "jianyiagi",
+  "license": "MIT"
+}