pixel-agents-vscode

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Pixel Agents VSCode Extension

Pixel Agents VS Code扩展

Skill by ara.so — AI Agent Skills collection.
Pixel Agents is a VS Code extension that visualizes multi-agent AI systems (currently Claude Code) as pixel art characters in a customizable office environment. Each agent becomes an animated character that reflects its real-time activity — typing when writing code, reading when searching files, and displaying speech bubbles when waiting for input.
ara.so开发的技能——AI Agent技能合集。
Pixel Agents是一款VS Code扩展,它可将多Agent AI系统(目前支持Claude Code)在可自定义的办公环境中以像素艺术角色的形式可视化展示。每个Agent都会变成一个动画角色,实时反映其活动状态——编写代码时会呈现打字动作,搜索文件时会呈现阅读动作,等待输入时会显示对话气泡。

Installation

安装

From VS Code Marketplace

从VS Code应用商店安装

  1. Open VS Code
  2. Go to Extensions (Ctrl+Shift+X / Cmd+Shift+X)
  3. Search for "Pixel Agents"
  4. Click Install
Or install via command line:
bash
code --install-extension pablodelucca.pixel-agents
  1. 打开VS Code
  2. 进入扩展面板(快捷键:Ctrl+Shift+X / Cmd+Shift+X)
  3. 搜索“Pixel Agents”
  4. 点击安装
或通过命令行安装:
bash
code --install-extension pablodelucca.pixel-agents

From Source

从源码安装

bash
git clone https://github.com/pablodelucca/pixel-agents.git
cd pixel-agents
npm install
cd webview-ui && npm install && cd ..
npm run build
Then press F5 in VS Code to launch the Extension Development Host.
bash
git clone https://github.com/pablodelucca/pixel-agents.git
cd pixel-agents
npm install
cd webview-ui && npm install && cd ..
npm run build
然后在VS Code中按F5启动扩展开发宿主环境。

Prerequisites

前置要求

  • VS Code: Version 1.105.0 or later
  • Claude Code CLI: Must be installed and configured
    bash
    # Install Claude Code (if not already installed)
    npm install -g @anthropic-ai/claude-code
  • VS Code:版本1.105.0或更高
  • Claude Code CLI:必须已安装并配置
    bash
    # 安装Claude Code(如未安装)
    npm install -g @anthropic-ai/claude-code

Key Features and Usage

核心功能与使用方法

Spawning Agents

创建Agent

Open the Pixel Agents panel (appears in bottom panel area alongside terminal):
Create normal agent:
typescript
// Click "+ Agent" button in Pixel Agents panel
// This spawns a new Claude Code terminal with a character
Create agent with skip permissions:
typescript
// Right-click "+ Agent" button
// Select "Launch with --dangerously-skip-permissions"
// Agent will bypass all tool approval prompts
打开Pixel Agents面板(位于底部面板区域,与终端并列):
创建普通Agent:
typescript
// 点击Pixel Agents面板中的“+ Agent”按钮
// 这会生成一个带有角色的新Claude Code终端
创建带有跳过权限的Agent:
typescript
// 右键点击“+ Agent”按钮
// 选择“Launch with --dangerously-skip-permissions”
// 该Agent将绕过所有工具授权提示

Managing Characters

管理角色

Assign character to a seat:
typescript
// 1. Click a character to select it
// 2. Click an empty seat to reassign the character
View agent status:
  • Walking: Agent is navigating to desk/idle
  • Typing: Writing code, creating files, making changes
  • Reading: Searching files, analyzing code
  • Speech bubble: Waiting for user input or permission
  • Idle: No current activity
为角色分配座位:
typescript
// 1. 点击一个角色选中它
// 2. 点击空座位重新分配角色
查看Agent状态:
  • 行走:Agent正在前往工位/处于空闲状态
  • 打字:正在编写代码、创建文件、修改内容
  • 阅读:正在搜索文件、分析代码
  • 对话气泡:正在等待用户输入或授权
  • 空闲:无当前活动

Office Layout Editor

办公布局编辑器

Open editor:
typescript
// Click "Layout" button in Pixel Agents panel
Editor tools:
  • Select (S): Select and move furniture
  • Paint (P): Paint floor tiles
  • Erase (E): Remove items
  • Place (F): Add furniture
  • Eyedropper (I): Pick colors/tiles
  • Pick (K): Select furniture type
Keyboard shortcuts:
  • Ctrl+Z
    /
    Cmd+Z
    : Undo (50 levels)
  • Ctrl+Y
    /
    Cmd+Y
    : Redo
  • Delete
    : Remove selected item
  • Arrow keys: Fine-tune placement
Expand grid:
typescript
// Click the ghost border outside current grid
// Grid expands up to 64×64 tiles
打开编辑器:
typescript
// 点击Pixel Agents面板中的“Layout”按钮
编辑器工具:
  • 选择(S):选择并移动家具
  • 绘制(P):绘制地板瓷砖
  • 擦除(E):移除物品
  • 放置(F):添加家具
  • 取色器(I):拾取颜色/瓷砖
  • 选择家具类型(K):选择要放置的家具类型
键盘快捷键:
  • Ctrl+Z
    /
    Cmd+Z
    :撤销(最多50级)
  • Ctrl+Y
    /
    Cmd+Y
    :重做
  • Delete
    :移除选中物品
  • 方向键:微调放置位置
扩展网格:
typescript
// 点击当前网格外的虚边框
// 网格最大可扩展至64×64瓷砖

Configuration

配置

Access settings via gear icon in Pixel Agents panel:
Sound notifications:
typescript
// Toggle sound when agent finishes turn
// Settings → Enable Sound Notifications
Debug view:
typescript
// Settings → Debug View
// Shows per-agent diagnostics:
// - JSONL file status
// - Lines parsed
// - Last data timestamp
// - File path
Export/Import layouts:
typescript
// Settings → Export Layout (saves as JSON)
// Settings → Import Layout (load from JSON file)
Add external assets:
typescript
// Settings → Add Asset Directory
// Point to folder with custom furniture packs
通过Pixel Agents面板中的齿轮图标访问设置:
声音通知:
typescript
// 切换Agent完成任务时的声音提示
// 设置 → Enable Sound Notifications
调试视图:
typescript
// 设置 → Debug View
// 显示每个Agent的诊断信息:
// - JSONL文件状态
// - 已解析行数
// - 最后数据时间戳
// - 文件路径
导出/导入布局:
typescript
// 设置 → Export Layout(保存为JSON格式)
// 设置 → Import Layout(从JSON文件加载)
添加外部资源:
typescript
// 设置 → Add Asset Directory
// 指向包含自定义家具包的文件夹

Working with Assets

资源使用说明

Asset Structure

资源结构

All assets are in
webview-ui/public/assets/
:
assets/
├── furniture/
│   ├── desk-01/
│   │   ├── manifest.json
│   │   └── sprite.png
│   └── chair-01/
│       ├── manifest.json
│       └── sprite.png
├── floors/
│   └── tile-wood.png
└── walls/
    └── brick/
        ├── manifest.json
        └── tileset.png
所有资源都位于
webview-ui/public/assets/
目录下:
assets/
├── furniture/
│   ├── desk-01/
│   │   ├── manifest.json
│   │   └── sprite.png
│   └── chair-01/
│       ├── manifest.json
│       └── sprite.png
├── floors/
│   └── tile-wood.png
└── walls/
    └── brick/
        ├── manifest.json
        └── tileset.png

Creating Custom Furniture

创建自定义家具

Manifest structure:
json
{
  "id": "desk-modern",
  "name": "Modern Desk",
  "category": "desks",
  "width": 2,
  "height": 1,
  "isSeat": true,
  "seatOffset": { "x": 0, "y": -16 },
  "rotationGroups": [
    {
      "rotations": ["north", "east", "south", "west"],
      "sprite": "desk-modern.png"
    }
  ],
  "stateGroups": [
    {
      "state": "off",
      "sprite": "desk-modern-off.png"
    },
    {
      "state": "on",
      "sprite": "desk-modern-on.png"
    }
  ]
}
Add to project:
bash
undefined
清单结构:
json
{
  "id": "desk-modern",
  "name": "Modern Desk",
  "category": "desks",
  "width": 2,
  "height": 1,
  "isSeat": true,
  "seatOffset": { "x": 0, "y": -16 },
  "rotationGroups": [
    {
      "rotations": ["north", "east", "south", "west"],
      "sprite": "desk-modern.png"
    }
  ],
  "stateGroups": [
    {
      "state": "off",
      "sprite": "desk-modern-off.png"
    },
    {
      "state": "on",
      "sprite": "desk-modern-on.png"
    }
  ]
}
添加到项目:
bash
undefined

1. Create folder in assets/furniture/

1. 在assets/furniture/目录下创建文件夹

mkdir webview-ui/public/assets/furniture/desk-modern
mkdir webview-ui/public/assets/furniture/desk-modern

2. Add sprite PNG and manifest.json

2. 添加精灵图PNG和manifest.json

3. Rebuild extension

3. 重新构建扩展

npm run build
undefined
npm run build
undefined

Using External Asset Directories

使用外部资源目录

typescript
// 1. Create asset directory structure:
// /my-assets/
//   furniture/
//     custom-desk/
//       manifest.json
//       sprite.png

// 2. In Pixel Agents Settings → Add Asset Directory
// 3. Select /my-assets/
// 4. Assets appear in furniture picker
External manifest.json:
json
{
  "id": "custom-plant",
  "name": "Custom Plant",
  "category": "decorations",
  "width": 1,
  "height": 1,
  "isSeat": false,
  "rotationGroups": [
    {
      "rotations": ["north"],
      "sprite": "plant.png"
    }
  ]
}
typescript
// 1. 创建资源目录结构:
// /my-assets/
//   furniture/
//     custom-desk/
//       manifest.json
//       sprite.png

// 2. 在Pixel Agents设置中 → Add Asset Directory
// 3. 选择/my-assets/
// 4. 资源将出现在家具选择器中
外部manifest.json示例:
json
{
  "id": "custom-plant",
  "name": "Custom Plant",
  "category": "decorations",
  "width": 1,
  "height": 1,
  "isSeat": false,
  "rotationGroups": [
    {
      "rotations": ["north"],
      "sprite": "plant.png"
    }
  ]
}

Code Examples

代码示例

Monitoring Agent Activity (Extension Side)

监控Agent活动(扩展端)

typescript
import * as vscode from 'vscode';
import * as fs from 'fs';
import * as path from 'path';

// Watch Claude Code JSONL transcript for agent activity
function watchAgentTranscript(projectPath: string, agentId: string): vscode.Disposable {
  const jsonlPath = path.join(
    projectPath,
    '.claude',
    'projects',
    agentId,
    'transcript.jsonl'
  );

  let lastPosition = 0;

  const interval = setInterval(() => {
    if (!fs.existsSync(jsonlPath)) {
      return;
    }

    const stats = fs.statSync(jsonlPath);
    if (stats.size <= lastPosition) {
      return;
    }

    const buffer = Buffer.alloc(stats.size - lastPosition);
    const fd = fs.openSync(jsonlPath, 'r');
    fs.readSync(fd, buffer, 0, buffer.length, lastPosition);
    fs.closeSync(fd);

    const lines = buffer.toString('utf-8').split('\n').filter(l => l.trim());
    
    lines.forEach(line => {
      try {
        const record = JSON.parse(line);
        handleAgentRecord(agentId, record);
      } catch (err) {
        console.error('[Pixel Agents] Failed to parse JSONL:', err);
      }
    });

    lastPosition = stats.size;
  }, 500);

  return new vscode.Disposable(() => clearInterval(interval));
}

function handleAgentRecord(agentId: string, record: any): void {
  // Detect agent activity from JSONL record type
  if (record.type === 'tool_use') {
    const toolName = record.content?.name;
    
    if (toolName === 'write_file' || toolName === 'edit_file') {
      updateAgentState(agentId, 'typing');
    } else if (toolName === 'search_files' || toolName === 'read_file') {
      updateAgentState(agentId, 'reading');
    } else if (toolName === 'run_command') {
      updateAgentState(agentId, 'typing');
    }
  } else if (record.type === 'user_message') {
    updateAgentState(agentId, 'idle');
  }
}

function updateAgentState(agentId: string, state: string): void {
  // Send message to webview
  webviewPanel.webview.postMessage({
    command: 'updateAgentState',
    agentId,
    state
  });
}
typescript
import * as vscode from 'vscode';
import * as fs from 'fs';
import * as path from 'path';

// 监控Claude Code JSONL记录文件以追踪Agent活动
function watchAgentTranscript(projectPath: string, agentId: string): vscode.Disposable {
  const jsonlPath = path.join(
    projectPath,
    '.claude',
    'projects',
    agentId,
    'transcript.jsonl'
  );

  let lastPosition = 0;

  const interval = setInterval(() => {
    if (!fs.existsSync(jsonlPath)) {
      return;
    }

    const stats = fs.statSync(jsonlPath);
    if (stats.size <= lastPosition) {
      return;
    }

    const buffer = Buffer.alloc(stats.size - lastPosition);
    const fd = fs.openSync(jsonlPath, 'r');
    fs.readSync(fd, buffer, 0, buffer.length, lastPosition);
    fs.closeSync(fd);

    const lines = buffer.toString('utf-8').split('\n').filter(l => l.trim());
    
    lines.forEach(line => {
      try {
        const record = JSON.parse(line);
        handleAgentRecord(agentId, record);
      } catch (err) {
        console.error('[Pixel Agents] Failed to parse JSONL:', err);
      }
    });

    lastPosition = stats.size;
  }, 500);

  return new vscode.Disposable(() => clearInterval(interval));
}

function handleAgentRecord(agentId: string, record: any): void {
  // 从JSONL记录类型检测Agent活动
  if (record.type === 'tool_use') {
    const toolName = record.content?.name;
    
    if (toolName === 'write_file' || toolName === 'edit_file') {
      updateAgentState(agentId, 'typing');
    } else if (toolName === 'search_files' || toolName === 'read_file') {
      updateAgentState(agentId, 'reading');
    } else if (toolName === 'run_command') {
      updateAgentState(agentId, 'typing');
    }
  } else if (record.type === 'user_message') {
    updateAgentState(agentId, 'idle');
  }
}

function updateAgentState(agentId: string, state: string): void {
  // 向Webview发送消息
  webviewPanel.webview.postMessage({
    command: 'updateAgentState',
    agentId,
    state
  });
}

Character Animation (Webview Side)

角色动画(Webview端)

typescript
interface Character {
  id: string;
  x: number;
  y: number;
  targetX: number;
  targetY: number;
  state: 'idle' | 'walking' | 'typing' | 'reading';
  direction: 'north' | 'south' | 'east' | 'west';
  frame: number;
  spriteSheet: HTMLImageElement;
}

class CharacterRenderer {
  private characters: Map<string, Character> = new Map();
  private readonly TILE_SIZE = 16;
  private readonly FRAME_RATE = 8; // frames per second
  private frameCounter = 0;

  update(deltaTime: number): void {
    this.frameCounter += deltaTime;
    const frameInterval = 1000 / this.FRAME_RATE;

    this.characters.forEach(char => {
      // Update position if walking
      if (char.state === 'walking') {
        const dx = char.targetX - char.x;
        const dy = char.targetY - char.y;
        const distance = Math.sqrt(dx * dx + dy * dy);

        if (distance < 1) {
          char.x = char.targetX;
          char.y = char.targetY;
          char.state = 'idle';
        } else {
          const speed = 2; // pixels per frame
          char.x += (dx / distance) * speed;
          char.y += (dy / distance) * speed;

          // Update direction based on movement
          if (Math.abs(dx) > Math.abs(dy)) {
            char.direction = dx > 0 ? 'east' : 'west';
          } else {
            char.direction = dy > 0 ? 'south' : 'north';
          }
        }
      }

      // Advance animation frame
      if (this.frameCounter >= frameInterval) {
        char.frame = (char.frame + 1) % this.getFrameCount(char.state);
      }
    });

    if (this.frameCounter >= frameInterval) {
      this.frameCounter = 0;
    }
  }

  render(ctx: CanvasRenderingContext2D): void {
    this.characters.forEach(char => {
      const spriteX = this.getSpriteX(char);
      const spriteY = this.getSpriteY(char);

      ctx.drawImage(
        char.spriteSheet,
        spriteX, spriteY,
        this.TILE_SIZE, this.TILE_SIZE * 2, // source
        Math.floor(char.x), Math.floor(char.y),
        this.TILE_SIZE, this.TILE_SIZE * 2  // destination
      );
    });
  }

  private getSpriteX(char: Character): number {
    return char.frame * this.TILE_SIZE;
  }

  private getSpriteY(char: Character): number {
    const directionOffsets = {
      'south': 0,
      'west': 1,
      'east': 2,
      'north': 3
    };
    
    const stateOffsets = {
      'idle': 0,
      'walking': 0,
      'typing': 4,
      'reading': 8
    };

    return (stateOffsets[char.state] + directionOffsets[char.direction]) 
           * this.TILE_SIZE * 2;
  }

  private getFrameCount(state: string): number {
    if (state === 'walking') return 4;
    if (state === 'typing' || state === 'reading') return 3;
    return 1; // idle
  }

  moveCharacter(id: string, targetX: number, targetY: number): void {
    const char = this.characters.get(id);
    if (char) {
      char.targetX = targetX;
      char.targetY = targetY;
      char.state = 'walking';
    }
  }

  setCharacterState(id: string, state: Character['state']): void {
    const char = this.characters.get(id);
    if (char) {
      char.state = state;
      char.frame = 0;
    }
  }
}
typescript
interface Character {
  id: string;
  x: number;
  y: number;
  targetX: number;
  targetY: number;
  state: 'idle' | 'walking' | 'typing' | 'reading';
  direction: 'north' | 'south' | 'east' | 'west';
  frame: number;
  spriteSheet: HTMLImageElement;
}

class CharacterRenderer {
  private characters: Map<string, Character> = new Map();
  private readonly TILE_SIZE = 16;
  private readonly FRAME_RATE = 8; // 每秒帧数
  private frameCounter = 0;

  update(deltaTime: number): void {
    this.frameCounter += deltaTime;
    const frameInterval = 1000 / this.FRAME_RATE;

    this.characters.forEach(char => {
      // 如果处于行走状态,更新位置
      if (char.state === 'walking') {
        const dx = char.targetX - char.x;
        const dy = char.targetY - char.y;
        const distance = Math.sqrt(dx * dx + dy * dy);

        if (distance < 1) {
          char.x = char.targetX;
          char.y = char.targetY;
          char.state = 'idle';
        } else {
          const speed = 2; // 每帧移动像素数
          char.x += (dx / distance) * speed;
          char.y += (dy / distance) * speed;

          // 根据移动方向更新角色朝向
          if (Math.abs(dx) > Math.abs(dy)) {
            char.direction = dx > 0 ? 'east' : 'west';
          } else {
            char.direction = dy > 0 ? 'south' : 'north';
          }
        }
      }

      // 更新动画帧
      if (this.frameCounter >= frameInterval) {
        char.frame = (char.frame + 1) % this.getFrameCount(char.state);
      }
    });

    if (this.frameCounter >= frameInterval) {
      this.frameCounter = 0;
    }
  }

  render(ctx: CanvasRenderingContext2D): void {
    this.characters.forEach(char => {
      const spriteX = this.getSpriteX(char);
      const spriteY = this.getSpriteY(char);

      ctx.drawImage(
        char.spriteSheet,
        spriteX, spriteY,
        this.TILE_SIZE, this.TILE_SIZE * 2, // 源区域
        Math.floor(char.x), Math.floor(char.y),
        this.TILE_SIZE, this.TILE_SIZE * 2  // 目标区域
      );
    });
  }

  private getSpriteX(char: Character): number {
    return char.frame * this.TILE_SIZE;
  }

  private getSpriteY(char: Character): number {
    const directionOffsets = {
      'south': 0,
      'west': 1,
      'east': 2,
      'north': 3
    };
    
    const stateOffsets = {
      'idle': 0,
      'walking': 0,
      'typing': 4,
      'reading': 8
    };

    return (stateOffsets[char.state] + directionOffsets[char.direction]) 
           * this.TILE_SIZE * 2;
  }

  private getFrameCount(state: string): number {
    if (state === 'walking') return 4;
    if (state === 'typing' || state === 'reading') return 3;
    return 1; // 空闲状态
  }

  moveCharacter(id: string, targetX: number, targetY: number): void {
    const char = this.characters.get(id);
    if (char) {
      char.targetX = targetX;
      char.targetY = targetY;
      char.state = 'walking';
    }
  }

  setCharacterState(id: string, state: Character['state']): void {
    const char = this.characters.get(id);
    if (char) {
      char.state = state;
      char.frame = 0;
    }
  }
}

Pathfinding (BFS for Agent Movement)

路径查找(Agent移动的BFS算法)

typescript
interface Point {
  x: number;
  y: number;
}

class Pathfinder {
  private grid: boolean[][]; // true = walkable
  private width: number;
  private height: number;

  constructor(width: number, height: number) {
    this.width = width;
    this.height = height;
    this.grid = Array(height).fill(null).map(() => Array(width).fill(true));
  }

  setObstacle(x: number, y: number, blocked: boolean): void {
    if (x >= 0 && x < this.width && y >= 0 && y < this.height) {
      this.grid[y][x] = !blocked;
    }
  }

  findPath(start: Point, end: Point): Point[] | null {
    if (!this.isWalkable(end.x, end.y)) return null;

    const queue: { point: Point; path: Point[] }[] = [
      { point: start, path: [start] }
    ];
    const visited = new Set<string>();
    visited.add(`${start.x},${start.y}`);

    const directions = [
      { x: 0, y: -1 }, // north
      { x: 1, y: 0 },  // east
      { x: 0, y: 1 },  // south
      { x: -1, y: 0 }  // west
    ];

    while (queue.length > 0) {
      const { point, path } = queue.shift()!;

      if (point.x === end.x && point.y === end.y) {
        return path;
      }

      for (const dir of directions) {
        const next = { x: point.x + dir.x, y: point.y + dir.y };
        const key = `${next.x},${next.y}`;

        if (
          this.isWalkable(next.x, next.y) &&
          !visited.has(key)
        ) {
          visited.add(key);
          queue.push({
            point: next,
            path: [...path, next]
          });
        }
      }
    }

    return null; // No path found
  }

  private isWalkable(x: number, y: number): boolean {
    return (
      x >= 0 &&
      x < this.width &&
      y >= 0 &&
      y < this.height &&
      this.grid[y][x]
    );
  }
}

// Usage
const pathfinder = new Pathfinder(32, 32);

// Mark furniture as obstacles
pathfinder.setObstacle(5, 5, true); // desk
pathfinder.setObstacle(6, 5, true); // desk continuation

// Find path from spawn to desk
const path = pathfinder.findPath({ x: 1, y: 1 }, { x: 5, y: 6 });

if (path) {
  // Move character along path
  path.forEach((point, index) => {
    setTimeout(() => {
      renderer.moveCharacter('agent-1', point.x * 16, point.y * 16);
    }, index * 200);
  });
}
typescript
interface Point {
  x: number;
  y: number;
}

class Pathfinder {
  private grid: boolean[][]; // true = 可通行
  private width: number;
  private height: number;

  constructor(width: number, height: number) {
    this.width = width;
    this.height = height;
    this.grid = Array(height).fill(null).map(() => Array(width).fill(true));
  }

  setObstacle(x: number, y: number, blocked: boolean): void {
    if (x >= 0 && x < this.width && y >= 0 && y < this.height) {
      this.grid[y][x] = !blocked;
    }
  }

  findPath(start: Point, end: Point): Point[] | null {
    if (!this.isWalkable(end.x, end.y)) return null;

    const queue: { point: Point; path: Point[] }[] = [
      { point: start, path: [start] }
    ];
    const visited = new Set<string>();
    visited.add(`${start.x},${start.y}`);

    const directions = [
      { x: 0, y: -1 }, // 北
      { x: 1, y: 0 },  // 东
      { x: 0, y: 1 },  // 南
      { x: -1, y: 0 }  // 西
    ];

    while (queue.length > 0) {
      const { point, path } = queue.shift()!;

      if (point.x === end.x && point.y === end.y) {
        return path;
      }

      for (const dir of directions) {
        const next = { x: point.x + dir.x, y: point.y + dir.y };
        const key = `${next.x},${next.y}`;

        if (
          this.isWalkable(next.x, next.y) &&
          !visited.has(key)
        ) {
          visited.add(key);
          queue.push({
            point: next,
            path: [...path, next]
          });
        }
      }
    }

    return null; // 未找到路径
  }

  private isWalkable(x: number, y: number): boolean {
    return (
      x >= 0 &&
      x < this.width &&
      y >= 0 &&
      y < this.height &&
      this.grid[y][x]
    );
  }
}

// 使用示例
const pathfinder = new Pathfinder(32, 32);

// 将家具标记为障碍物
pathfinder.setObstacle(5, 5, true); // 桌子
pathfinder.setObstacle(6, 5, true); // 桌子延伸部分

// 查找从出生点到桌子的路径
const path = pathfinder.findPath({ x: 1, y: 1 }, { x: 5, y: 6 });

if (path) {
  // 沿路径移动角色
  path.forEach((point, index) => {
    setTimeout(() => {
      renderer.moveCharacter('agent-1', point.x * 16, point.y * 16);
    }, index * 200);
  });
}

Troubleshooting

故障排除

Agent Not Appearing

Agent未显示

Enable Debug View:
typescript
// Settings → Debug View (toggle on)
// Check per-agent diagnostics:
// - "JSONL not found" = extension can't locate session file
// - "Lines parsed: 0" = no activity detected
// - Verify file path matches expected location
Check Debug Console:
typescript
// If running from source (F5 Development Host):
// View → Debug Console
// Search for "[Pixel Agents]"
// Look for:
// - Project directory resolution errors
// - JSONL polling status
// - Path encoding mismatches
启用调试视图:
typescript
// 设置 → Debug View(开启)
// 查看每个Agent的诊断信息:
// - "JSONL not found" = 扩展无法找到会话文件
// - "Lines parsed: 0" = 未检测到活动
// - 验证文件路径是否与预期位置匹配
检查调试控制台:
typescript
// 如果从源码运行(F5开发宿主环境):
// 视图 → 调试控制台
// 搜索"[Pixel Agents]"
// 查找以下信息:
// - 项目目录解析错误
// - JSONL轮询状态
// - 路径编码不匹配

Agent Stuck on Idle

Agent一直处于空闲状态

Common causes:
  1. JSONL file not found: Claude Code session hasn't created transcript yet
    bash
    # Check if file exists
    ls ~/.claude/projects/*/transcript.jsonl
  2. Polling lag: Extension polls every 500ms, brief delays are normal
  3. Unrecognized record type: Claude Code may emit new JSONL types
    typescript
    // Check Debug Console for "Unrecognized JSONL record type: X"
常见原因:
  1. JSONL文件未找到:Claude Code会话尚未创建记录文件
    bash
    # 检查文件是否存在
    ls ~/.claude/projects/*/transcript.jsonl
  2. 轮询延迟:扩展每500ms轮询一次,短暂延迟属于正常现象
  3. 无法识别的记录类型:Claude Code可能会生成新的JSONL类型
    typescript
    // 在调试控制台中查找"Unrecognized JSONL record type: X"

Agent-Terminal Desync

Agent与终端不同步

Symptoms:
  • Character appears but terminal is closed
  • Terminal exists but no character
  • Multiple characters for one terminal
Fixes:
typescript
// 1. Close all Claude Code terminals
// 2. In Pixel Agents panel, click each character
// 3. Press Delete or click trash icon
// 4. Spawn fresh agent with "+ Agent"
症状:
  • 角色已显示但终端已关闭
  • 终端存在但无角色显示
  • 一个终端对应多个角色
修复方法:
typescript
// 1. 关闭所有Claude Code终端
// 2. 在Pixel Agents面板中,点击每个角色
// 3. 按Delete键或点击垃圾桶图标
// 4. 点击"+ Agent"重新创建Agent

Speech Bubble Not Clearing

对话气泡无法清除

Heuristic-based detection limitations:
The extension uses idle timers to detect when agents wait for input. If speech bubble persists:
typescript
// 1. Type something in the Claude Code terminal
// 2. Wait 2-3 seconds for polling to update
// 3. If stuck, close and reopen Pixel Agents panel
基于启发式检测的局限性:
扩展使用空闲计时器检测Agent是否等待输入。如果对话气泡持续显示:
typescript
// 1. 在Claude Code终端中输入内容
// 2. 等待2-3秒,待轮询更新状态
// 3. 如果仍未清除,关闭并重新打开Pixel Agents面板

Linux/macOS Home Directory Confusion

Linux/macOS主目录识别问题

If you launch VS Code without opening a folder:
bash
undefined
如果未打开文件夹就启动VS Code:
bash
undefined

Agent starts in home directory

Agent将在主目录启动

code
code

Sessions tracked under:

会话记录存储在:

~/.claude/projects/<home-directory-hash>/

**Solution:**

```bash
~/.claude/projects/<home-directory-hash>/

**解决方案:**

```bash

Always open VS Code with a folder

始终打开VS Code时指定文件夹

code /path/to/project
undefined
code /path/to/project
undefined

Common Patterns

常见使用模式

Multi-Agent Workflow

多Agent工作流

typescript
// 1. Spawn multiple agents for parallel work
// Right-click "+ Agent" → create 3 agents

// 2. Assign each to different task
// Agent 1: "Refactor authentication module"
// Agent 2: "Write unit tests for API"
// Agent 3: "Update documentation"

// 3. Monitor all agents visually in office
// Watch characters move between typing/reading states

// 4. Intervene when speech bubble appears
// Click terminal to provide input or approval
typescript
// 1. 创建多个Agent并行工作
// 右键点击"+ Agent" → 创建3个Agent

// 2. 为每个Agent分配不同任务
// Agent 1:"重构认证模块"
// Agent 2:"为API编写单元测试"
// Agent 3:"更新文档"

// 3. 在办公界面中可视化监控所有Agent
// 观察角色在打字/阅读状态间切换

// 4. 当对话气泡出现时进行干预
// 点击终端提供输入或授权

Custom Office for Team Visualization

用于团队可视化的自定义办公布局

typescript
// 1. Design office layout per team structure
// Layout → Create 4 desk clusters for 4 teams

// 2. Export layout for sharing
// Settings → Export Layout → save as team-office.json

// 3. Team members import same layout
// Settings → Import Layout → select team-office.json

// 4. Everyone sees agents in same office structure
typescript
// 1. 根据团队结构设计办公布局
// 布局 → 创建4个工位集群对应4个团队

// 2. 导出布局以便共享
// 设置 → Export Layout → 保存为team-office.json

// 3. 团队成员导入相同布局
// 设置 → Import Layout → 选择team-office.json

// 4. 所有人将在相同的办公结构中查看Agent

Sub-Agent Visualization

子Agent可视化

When Claude Code uses the
task
tool to spawn sub-agents:
typescript
// Parent agent spawns sub-agent
// Example: Main agent delegates "write tests" to sub-agent

// Visual behavior:
// - New character appears linked to parent
// - Sub-agent character animates independently
// - Both visible until sub-task completes
// - Sub-agent disappears when task done
当Claude Code使用
task
工具创建子Agent时:
typescript
// 父Agent创建子Agent
// 示例:主Agent将"编写测试"任务委托给子Agent

// 可视化表现:
// - 新角色出现并关联到父Agent
// - 子Agent角色独立动画
// - 两者将同时显示直到子任务完成
// - 子任务完成后子Agent消失

External Asset Workflow

外部资源工作流

typescript
// 1. Create asset pack structure
mkdir -p /my-assets/furniture/custom-furniture

// 2. Add manifest and sprites
cat > /my-assets/furniture/custom-furniture/manifest.json << 'EOF'
{
  "id": "gaming-chair",
  "name": "Gaming Chair",
  "category": "seats",
  "width": 1,
  "height": 1,
  "isSeat": true,
  "seatOffset": { "x": 0, "y": -8 },
  "rotationGroups": [
    {
      "rotations": ["north", "south", "east", "west"],
      "sprite": "chair.png"
    }
  ]
}
EOF

// 3. Link in Pixel Agents
// Settings → Add Asset Directory → /my-assets

// 4. Use in layout editor
// Layout → Place → Select "Gaming Chair"
typescript
// 1. 创建资源包结构
mkdir -p /my-assets/furniture/custom-furniture

// 2. 添加清单和精灵图
cat > /my-assets/furniture/custom-furniture/manifest.json << 'EOF'
{
  "id": "gaming-chair",
  "name": "Gaming Chair",
  "category": "seats",
  "width": 1,
  "height": 1,
  "isSeat": true,
  "seatOffset": { "x": 0, "y": -8 },
  "rotationGroups": [
    {
      "rotations": ["north", "south", "east", "west"],
      "sprite": "chair.png"
    }
  ]
}
EOF

// 3. 在Pixel Agents中关联
// 设置 → Add Asset Directory → /my-assets

// 4. 在布局编辑器中使用
// 布局 → Place → 选择"Gaming Chair"

Extension Commands

扩展命令

Pixel Agents registers the following VS Code commands:
typescript
// Open Pixel Agents panel
"pixelAgents.openPanel"

// Spawn new agent (normal)
"pixelAgents.spawnAgent"

// Spawn agent with skip permissions
"pixelAgents.spawnAgentSkipPermissions"

// Open layout editor
"pixelAgents.openLayoutEditor"

// Open settings modal
"pixelAgents.openSettings"

// Toggle debug view
"pixelAgents.toggleDebug"
Access via Command Palette (
Ctrl+Shift+P
/
Cmd+Shift+P
):
> Pixel Agents: Open Panel
> Pixel Agents: Spawn Agent
Pixel Agents注册了以下VS Code命令:
typescript
// 打开Pixel Agents面板
"pixelAgents.openPanel"

// 创建新Agent(普通模式)
"pixelAgents.spawnAgent"

// 创建带有跳过权限的Agent
"pixelAgents.spawnAgentSkipPermissions"

// 打开布局编辑器
"pixelAgents.openLayoutEditor"

// 打开设置窗口
"pixelAgents.openSettings"

// 切换调试视图
"pixelAgents.toggleDebug"
通过命令面板访问(快捷键:
Ctrl+Shift+P
/
Cmd+Shift+P
):
> Pixel Agents: Open Panel
> Pixel Agents: Spawn Agent

Performance Considerations

性能考量

JSONL polling:
  • Polls every 500ms per agent
  • Minimal CPU impact for <10 agents
  • For 10+ agents, consider increasing interval in source
Canvas rendering:
  • 60 FPS target, degrades gracefully
  • Tested up to 64×64 grid with 20 characters
  • Use integer zoom levels for pixel-perfect rendering
Memory usage:
  • ~5-10 MB per agent (JSONL tracking + character state)
  • Asset cache: ~20-50 MB depending on furniture count
JSONL轮询:
  • 每个Agent每500ms轮询一次
  • 当Agent数量少于10个时,CPU影响极小
  • 当Agent数量超过10个时,可考虑在源码中增加轮询间隔
Canvas渲染:
  • 目标帧率为60FPS,可优雅降级
  • 已测试64×64网格搭配20个角色的场景
  • 使用整数缩放级别以实现像素完美渲染
内存使用:
  • 每个Agent约占用5-10MB(JSONL追踪 + 角色状态)
  • 资源缓存:根据家具数量约占用20-50MB

Future Roadmap

未来规划

Pixel Agents is evolving toward a fully agent-agnostic, platform-agnostic interface:
  • Agent-agnostic adapters: Support for Codex, Cursor, Copilot, Gemini
  • Platform flexibility: Electron app, web app, beyond VS Code
  • Advanced features:
    • Desks as directories (drag agent to desk = assign to project)
    • Kanban board for autonomous task picking
    • Deep agent inspection (model, context, history, prompts)
    • Token health bars (rate limits, context windows)
    • 3D/VR office environments
See GitHub Discussions for ongoing architecture work.
Pixel Agents正朝着完全Agent无关、平台无关的界面方向发展:
  • Agent适配层:支持Codex、Cursor、Copilot、Gemini
  • 平台灵活性:Electron应用、Web应用,超越VS Code限制
  • 高级功能
    • 工位对应目录(将Agent拖到工位 = 分配到项目)
    • 用于自主任务选择的看板
    • Agent深度检查(模型、上下文、历史记录、提示词)
    • Token健康条(速率限制、上下文窗口)
    • 3D/VR办公环境
查看GitHub Discussions了解正在进行的架构工作。