lonny-agent 0.2.4 → 0.2.7

package/.lonny/plan-web-cwd-status.md

@@ -1,38 +0,0 @@
-# Plan: Web UI 状态栏添加工作目录显示
-
-## 修改文件
-
-### 1. `src/web/index.ts` — 后端
-在 `hello` 消息中增加 `cwd` 字段：
-```ts
-cwd: config.cwd,
-```
-添加在 `version` 之后。
-
-### 2. `src/web/public/index.html` — HTML
-在状态栏左侧（`<div class="status-left">` 内）添加：
-```html
-<span class="separator">|</span>
-<span id="cwd-display" class="cwd-text"></span>
-```
-放在 `mode-display` 和 `</div>` 之间。
-
-### 3. `src/web/public/app.js` — 前端 JS
-- 添加 DOM 引用：`const cwdDisplay = document.getElementById('cwd-display')`
-- 在 `hello` 消息处理中添加：
-```js
-cwdDisplay.textContent = msg.cwd || ''
-```
-
-### 4. `src/web/public/style.css` — 样式
-添加：
-```css
-.cwd-text { color: var(--text-dim); font-size: 12px; max-width: 200px; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; vertical-align: middle; }
-```
-
-## Todo List
-- [x] `src/web/index.ts` — hello 消息添加 `cwd: config.cwd`
-- [x] `src/web/public/index.html` — 状态栏添加 cwd 元素
-- [x] `src/web/public/app.js` — 添加 DOM 引用和消息处理
-- [x] `src/web/public/style.css` — 添加 cwd 样式
-- [x] 构建并验证

package/AGENT.md

@@ -1,99 +0,0 @@
-# CLAUDE.md — lonny-agent
-
-This file provides guidance to AI coding agents working with this repository.
-
-## Project Overview
-
-**lonny-agent** is a per-call-pricing-optimized AI coding assistant that runs in the terminal. It supports three modes (`code`, `plan`, `ask`), multiple LLM providers (Anthropic, OpenAI, Google, Ollama), and features a TUI built with `@earendil-works/pi-tui`.
-
-### Key Architecture
-
-```
-src/
-├── agent/          # Core agent loop, session management, LLM providers
-│   ├── session.ts  # Main session loop (message handling, tool dispatch)
-│   ├── llm.ts      # LLM provider interface
-│   ├── event-bus.ts # Decoupled event system for TUI ↔ agent communication
-│   ├── compaction.ts # Context compaction (message history summarization)
-│   ├── prompt-builder.ts # System prompt construction
-│   └── providers/  # Anthropic, OpenAI, Google, Ollama
-├── cli/            # CLI argument parsing (commander)
-├── config/         # Configuration loading (env, file, defaults)
-├── diff/           # File read tracking and edit validation
-├── tools/          # Tool implementations (read, edit, bash, glob, grep, etc.)
-│   ├── registry.ts # ToolRegistry with plugin support
-│   ├── exec.ts     # JavaScript sandbox executor (inspired by Codex CLI)
-│   └── types.ts    # Tool interface definitions
-├── tui/            # Terminal UI (landing page, chat display, status bar)
-└── index.ts        # Entry point
-```
-
-### Core Design Decisions
-
-1. **Per-call cost optimization**: System prompt tells the model to batch operations into as few API calls as possible
-2. **Three modes**: `code` (direct editing), `plan` (analysis + plan document), `ask` (Q&A only)
-3. **Event bus decoupling**: TUI and agent communicate through an event bus, not direct method calls
-4. **Exec sandbox**: JavaScript VM sandbox lets the model orchestrate multi-step operations in one API call
-5. **File read tracking**: Prevents editing files that haven't been read or were modified externally
-6. **Context compaction**: Summarizes older conversation turns when token budget is exceeded
-
-## Git Commit Convention
-
-Use **Conventional Commits**:
-
-```
-<type>: <description>
-```
-
-Types: `feat`, `fix`, `docs`, `chore`, `refactor`, `test`
-
-Examples:
-- `feat: add Tavily web search tool`
-- `fix: handle CRLF line endings in edit tool`
-- `chore: add Biome + husky code quality pipeline`
-
-## Commands
-
-```bash
-npm install        # Install dependencies
-npm run dev        # Dev mode (tsx watch)
-npm run build      # Build (tsc)
-npm start          # Run built version
-npm test           # Run tests (vitest)
-npm test:watch     # Watch mode tests
-
-# Lint & Format (Biome)
-npm run lint       # Check only
-npm run lint:fix   # Auto-fix
-npm run format     # Format all
-npm run check      # Lint + format check
-npm run check:fix  # Lint + format auto-fix
-```
-
-## Code Quality
-
-- **Formatter**: Biome (2-space indent, 100 line width, single quotes, as-needed semicolons)
-- **Pre-commit**: husky + lint-staged (auto-fix staged files with Biome)
-- **TypeScript**: strict mode, no unchecked index access
-- **Testing**: vitest with `.test.ts` files co-located in `__tests__/` directories
-
-## Architecture Notes
-
-### Tool System
-- `ToolRegistry` (registry.ts) manages tool registration and dispatch
-- Tools implement `Tool` interface (definition + execute method)
-- Plugin system allows external tool registration via `registerPlugin()`
-- `exec` tool exposes all registered tools as `await tools.xxx(args)` in a sandboxed VM
-
-### Session Flow
-1. User sends message → `Session.chat()`
-2. System prompt + message history + tool definitions → LLM provider
-3. LLM streams response (text + tool calls)
-4. Tool calls dispatched via `ToolRegistry.dispatch()`
-5. Results appended to message history
-6. Compaction triggered if token budget exceeded
-
-### Testing
-- Unit tests: `src/**/__tests__/*.test.ts`
-- Mock strategy: mock only modules with side effects (network, filesystem)
-- Avoid mocking pure functions or data modules

package/src/tools/bash.ts

@@ -1,39 +0,0 @@
-import { exec } from 'node:child_process'
-import { promisify } from 'node:util'
-
-const execAsync = promisify(exec)
-
-import { fmtErr } from './errors.js'
-import type { Tool, ToolResult } from './types.js'
-
-export const bashTool: Tool = {
-  definition: {
-    name: 'bash',
-    description: 'Execute a shell command. Returns stdout and stderr.',
-    parameters: {
-      command: { type: 'string', description: 'Shell command to execute', required: true },
-      description: { type: 'string', description: 'Brief description of what the command does' },
-      timeout: { type: 'number', description: 'Timeout in milliseconds (default: 120000)' },
-    },
-  },
-  async execute(input): Promise<ToolResult> {
-    const command = input.command as string
-    if (!command) {
-      return { success: false, output: '', error: 'command is required' }
-    }
-
-    const timeout = (input.timeout as number) || 120_000
-
-    try {
-      const { stdout } = await execAsync(command, {
-        encoding: 'utf-8',
-        timeout,
-        maxBuffer: 10 * 1024 * 1024,
-      })
-      return { success: true, output: stdout || '(no output)' }
-    } catch (err) {
-      const msg = fmtErr(err)
-      return { success: false, output: '', error: `Command failed: ${msg}` }
-    }
-  },
-}