shuiyu67
b577b0ff21
方案 A（去重）: - mcp_to_openai_tools 末尾按 function.name 去重，消除 cache+builtin 叠加 造成的同名重复（memory/scheduler/lsp__* 等），FC 工具 42→36，去掉 约 14% 的垃圾 schema token - get_liugin_usage_prompts 的 plugins 段由逐条 name:desc 改为工具名清单， 消除与 FC tools 字段的双重曝光，保留无 FC 模型的回退清单 方案 B（分层裁剪）: - 新增 prune_tools(fc_tools, user_input): 常驻 ALWAYS(tool_search/memory/ scheduler) + 外部桥接(mcp__)默认保留 + 按关键词信号追加对应工具 (代码/文件->code_editor+file_manager+lsp__*; git/shell/浏览器/搜索/子agent/ 工程化各按信号) - cli_core 两处 FC 构建点(TUI/CLI)接入 prune_tools(current_input) 实测(AICLI): 纯聊天 FC token ~4310->285(降93%)，写代码 ~973(降77%) 测试: 新增 tests/test_prune_tools.py (7 passed)，全量 324 passed 无回归
XiaoLi Pro-CLI
AI Programming Assistant — Code Editing · Git · Browser Automation · Multi-Agent
Features
- Dual AI Engines — Ollama local + OpenAI compatible (DeepSeek, Grok, SiliconFlow, etc.)
- Code Editing & Search — Precise replace, batch edit, symbol extraction, regex search
- Git Integration — Full Git operations + workflow automation
- Browser Automation — Playwright-driven, navigate/interact/screenshot/JS execution
- Sub-Agent System — Multi-agent parallel collaboration with task delegation
- Unified Safety — AI risk analysis + 3-mode switching (Normal/Manual/Unrestricted)
- 15 Plugins, 141 Operations — Code, Git, browser, engineering, file management, etc.
- 113 Automated Tests — pytest coverage for all plugins and core modules
Quick Start
git clone https://gitee.com/shuiyu1123/xiaoli-cli.git
cd xiaoli-cli
pip install -r requirements.txt
python ai_cli.py
AI Engines
| Engine | Type | Description |
|---|---|---|
ollama |
Local | Ollama local models, no API key needed |
openai |
Cloud/Local | OpenAI compatible: DeepSeek, Grok, SiliconFlow, local vLLM |
Plugins (15)
Core Productivity
| Plugin | Ops | Description |
|---|---|---|
code_editor |
23 | Code editing + search. replace/batch/find/regex/symbols/imports/callers/todo |
git_tools |
20 | Git. status/diff/log/commit/branch + smart-commit/changelog/contributors |
cmd_executor |
1 | Shell commands. Cross-platform, timeout, danger blocking |
file_manager |
11 | File system. list/read/write/copy/move/delete/search/info/mkdir |
auto_engineer |
16 | Engineering. lint/format/test/build/deps + complexity/security/metrics |
task_manager |
7 | Task tracking. add/list/complete/delete/clear with JSON persistence |
Advanced
| Plugin | Ops | Description |
|---|---|---|
sub_agent |
8 | Sub-Agent system. spawn/list/status/result/send/kill, parallel collaboration |
browser_auto |
31 | Browser automation. navigate/click/fill/screenshot/JS/PDF/cookies |
tool_search |
— | Tool discovery. Search available plugins by keyword |
Utilities
| Plugin | Description |
|---|---|
network_tools |
ping/get/status/headers/ip |
ai_search |
AI search via jina.ai + web content extraction |
frontend_tester |
HTML/CSS/JS syntax check, responsive validation |
speech_recognition |
Voice recording + speech-to-text |
audio_player |
Audio playback |
send_image |
Send images to phone (Clawli mode) |
Safety
Three modes (/safe to switch):
| Mode | Icon | Behavior |
|---|---|---|
| Normal (default) | AI analyzes risk, confirms only when risk detected | |
| Manual | All commands require user confirmation | |
| Unrestricted | Execute directly, no checking |
Testing
python -m pytest tests/ -v # 113 tests