Releases: lyy0709/ContextWeaver
Releases · lyy0709/ContextWeaver
v1.0.6
What's Changed
Prompt Enhancer: Ghostwriter Style Rewrite
Replaced the planning-oriented prompt enhancement template with a ghostwriter/optimizer approach:
- Before: Enhanced prompts would expand into detailed task plans with acceptance criteria, step breakdowns, and boundary conditions
- After: Enhanced prompts stay concise — disambiguates wording, leverages conversation history and codebase context to fill in concrete references, without inventing new requirements
Key changes
- Bilingual templates (EN/ZH) auto-selected by detected input language
- System prompt identity changed from "enhancement assistant" to "prompt ghostwriter and optimizer"
- First-person voice ("I") for rewritten prompts
- Removed
{{language_instruction}}placeholder (language is now determined by template selection)
Full Changelog: v1.0.5...v1.0.6
v1.0.5
Prompt Enhancer Hardening & Optimization
New Features
- Retry with backoff: All LLM adapters now have timeout (90s), exponential backoff retry for 429/5xx errors
- Configurable Web UI timeout:
PROMPT_ENHANCER_WEBUI_TIMEOUT_MSenv var (default 8min, clamped 30s–60min) - Smart technical terms extraction: Heuristic extraction of identifiers/paths from prompt for better codebase retrieval
Improvements
- Strict parser fallback: Returns original prompt instead of raw LLM output on parse failure
- Markdown fence stripping: Handles models that wrap output in ```xml fences
- Stronger system prompt: Format-locked with
<cw-end/>sentinel stop sequences across all endpoints - Lower temperature: 0.7 → 0.3 for more deterministic output
- Conditional template blocks: Empty conversation_history/codebase_context sections are fully omitted (saves tokens)
- Context budget: Retrieval budget tightened from 48k to 12k chars for enhancer use case
- Config DRY: Default endpoint URLs/models consolidated into single
ENHANCER_DEFAULTSconstant - Robust error parsing: Adapters handle non-JSON error responses gracefully
- Response body cleanup: Socket resources released before retry
Testing
- 101 tests passing (up from 75)
- New test suites: fetchWithRetry, extractTechnicalTerms, expanded parser/template/ui tests
v1.0.4
What's Changed
MCP enhance-prompt Web UI 模式
- 新增
webui参数:MCPenhance-prompt工具新增可选布尔参数webui(默认true)webui=true(默认):自动打开浏览器 Web UI,用户可交互式审阅、编辑、重新增强 prompt 后返回结果webui=false:保持原有行为,直接通过 API 返回增强结果
- 完整上下文透传:
startEnhanceServer()现在正确传递conversationHistory和projectRootPath到 LLM 调用
Full Changelog: v1.0.3...v1.0.4
v1.0.3
What's New
Prompt Enhancer Web UI
- 黑白配色 — 全新纯黑白色系设计
- 中文界面 — 所有 UI 文本改为中文
- 8 分钟倒计时 — 可视化倒计时,超时自动采用增强版结果
- 响应式布局优化 — flex 弹性 textarea,适配各种屏幕尺寸
Bug Fixes
- 修复 DTS 构建报错(
src/mcp/main.ts缺少export {})
Docs
- 新增
README.en.md英文文档 - README 添加 Prompt Enhancer Web UI 截图
Install
npm install -g @lyy0709/contextweaverFull Changelog: v1.0.0...v1.0.3
v1.0.0 — Prompt Enhancer
🚀 ContextWeaver v1.0.0 — Prompt Enhancer
基于 hsingjui/ContextWeaver 的增强版本,新增 Prompt Enhancer(提示词增强) 功能。
✨ 新增功能
Prompt Enhancer — 多 LLM 提示词增强
- MCP 工具
enhance-prompt:AI 助手可直接调用,增强用户提示词 - CLI 命令
cw enhance "prompt":支持--no-browser直接输出模式 - Web UI 交互:本地浏览器编辑界面,支持重新增强、使用原始文本、手动编辑
- 多 API 端点:支持 OpenAI / Claude / Gemini,通过环境变量一键切换
- 自动语言检测:中文输入自动用中文回复
- 可自定义模板:支持自定义增强提示词模板
📦 安装
# npm
npm install -g @lyy0709/contextweaver
# pnpm
pnpm add -g @lyy0709/contextweaver⚙️ 配置 Prompt Enhancer
编辑 ~/.contextweaver/.env,添加:
PROMPT_ENHANCER_ENDPOINT=openai # openai / claude / gemini
PROMPT_ENHANCER_TOKEN=your-api-key # 必填
# PROMPT_ENHANCER_MODEL= # 可选,覆盖默认模型🧪 测试
新增 Vitest 测试框架,79 个单元测试覆盖所有 enhancer 模块。
pnpm test # 运行测试📝 完整变更
- 新增
src/enhancer/模块(11 个文件) - 新增
src/mcp/tools/enhancePrompt.tsMCP 工具 - 更新
src/config.ts增加 EnhancerConfig - 更新
src/index.ts增加enhanceCLI 命令 - 更新
src/mcp/server.ts注册 enhance-prompt 工具 - 新增
tests/目录(10 个测试文件,79 个用例) - 新增
vitest.config.ts