task-director
Task Director — turn complex tasks into movie storyboards. Create a plan, review it, then execute step by step with fallback support. Pause, retry, skip anytime.
Best use case
task-director is best used when you need a repeatable AI agent workflow instead of a one-off prompt.
Task Director — turn complex tasks into movie storyboards. Create a plan, review it, then execute step by step with fallback support. Pause, retry, skip anytime.
Teams using task-director should expect a more consistent output, faster repeated execution, less prompt rewriting.
When to use this skill
- You want a reusable workflow that can be run more than once with consistent structure.
When not to use this skill
- You only need a quick one-off answer and do not need a reusable workflow.
- You cannot install or maintain the underlying files, dependencies, or repository context.
Installation
Claude Code / Cursor / Codex
Manual Installation
- Download SKILL.md from GitHub
- Place it in
.claude/skills/task-director/SKILL.mdinside your project - Restart your AI agent — it will auto-discover the skill
How task-director Compares
| Feature / Agent | task-director | Standard Approach |
|---|---|---|
| Platform Support | Not specified | Limited / Varies |
| Context Awareness | High | Baseline |
| Installation Complexity | Unknown | N/A |
Frequently Asked Questions
What does this skill do?
Task Director — turn complex tasks into movie storyboards. Create a plan, review it, then execute step by step with fallback support. Pause, retry, skip anytime.
Where can I find the source code?
You can find the source code on GitHub using the link provided at the top of the page.
Related Guides
Top AI Agents for Productivity
See the top AI agent skills for productivity, workflow automation, operational systems, documentation, and everyday task execution.
AI Agents for Coding
Browse AI agent skills for coding, debugging, testing, refactoring, code review, and developer workflows across Claude, Cursor, and Codex.
Best AI Skills for Claude
Explore the best AI skills for Claude and Claude Code across coding, research, workflow automation, documentation, and agent operations.
SKILL.md Source
# 任务导演 / Task Director
把复杂任务编排成电影剧本。**先看分镜,确认后再开机。**
No more "run and pray". Plan first, execute with confidence.
## 核心理念
普通 Agent 的工作方式:
```
用户:部署这个项目
Agent:冲!(5分钟后报错)啊... 从头再来...
```
用任务导演的 Agent:
```
用户:部署这个项目
Agent:我给你写了个分镜,你看看?
🎬 第一幕:环境检查(预计 30s)
🎬 第二幕:安装依赖(预计 2min)
🎬 第三幕:构建部署(预计 5min)
🎭 备选:先本地测试
🎬 杀青:验证服务能访问
用户:第三幕的备选方案换成本地构建
Agent:好的,改了。开机?
用户:开机!
Agent:🎬 Action! 第一幕...
```
## 命令参考
DIRECTOR = python <skill_dir>/scripts/director.py
### 创建剧本
```bash
$DIRECTOR create --title "部署全栈项目" --scenes scenes.json
```
scenes.json 格式:
```json
[
{
"name": "环境准备",
"description": "检查并配置运行环境",
"estimated_time": "1 min",
"shots": [
{
"action": "检查 Node.js 版本",
"command": "node --version"
},
{
"action": "安装项目依赖",
"command": "npm install",
"fallback": {
"action": "使用 yarn 安装",
"command": "yarn install"
}
}
]
},
{
"name": "构建部署",
"shots": [
{
"action": "执行构建",
"command": "npm run build"
}
]
}
]
```
**Shot 的字段:**
- `action` (必填): 要做什么
- `command` (可选): 具体命令
- `fallback` (可选): 失败后的备选方案
### 查看剧本
```bash
$DIRECTOR show # 显示当前活跃的剧本
$DIRECTOR show --id movie_xxx # 按 ID 查看
```
### 批准剧本
```bash
$DIRECTOR approve
```
状态从 `draft` → `approved`,可以开始执行了。
### 开机拍摄(执行下一步)
```bash
$DIRECTOR action
```
输出下一个待执行的 shot 信息(JSON 格式),Agent 根据这个信息去实际执行。
指定某一幕某个镜头:
```bash
$DIRECTOR action --scene 2 --shot 1
```
### 记录结果
Agent 执行完后,回报结果:
```bash
# 成功
$DIRECTOR result --id movie_xxx --outcome done --output "v18.0.0"
# 失败
$DIRECTOR result --id movie_xxx --outcome ng --output "npm ERR! code ERESOLVE"
# 用备选方案成功
$DIRECTOR result --id movie_xxx --outcome done --fallback --output "yarn install success"
```
### 暂停 / 继续
```bash
$DIRECTOR cut # 暂停(导演喊卡)
$DIRECTOR action # 继续拍摄
```
### 跳过镜头
```bash
$DIRECTOR skip --scene 1 --shot 3 --reason "不需要这一步"
```
### 杀青
```bash
$DIRECTOR wrap
```
显示最终统计:耗时、完成率、失败数。
### 查看所有剧本
```bash
$DIRECTOR list
```
## Agent 工作流
### 创建剧本(Agent 自动生成)
```python
import json
scenes = [
{
"name": "环境检查",
"estimated_time": "30s",
"shots": [
{"action": "检查 Python 版本", "command": "python --version"},
{"action": "检查依赖是否安装", "command": "pip list | grep flask"},
]
},
{
"name": "启动服务",
"estimated_time": "10s",
"shots": [
{
"action": "启动 Flask 服务",
"command": "python app.py",
"fallback": {
"action": "使用 gunicorn 启动",
"command": "gunicorn app:app"
}
},
]
},
{
"name": "验证",
"estimated_time": "5s",
"shots": [
{"action": "检查端口是否监听", "command": "curl http://localhost:5000/health"},
]
},
]
with open("scenes.json", "w") as f:
json.dump(scenes, f, ensure_ascii=False, indent=2)
```
### 执行循环
```
1. $DIRECTOR action → 获取下一个 shot
2. Agent 执行 shot 中的 command
3. $DIRECTOR result --id xxx --outcome done --output "..." → 记录结果
4. 如果有下一个,回到步骤 1
5. 全部完成后 $DIRECTOR wrap → 杀青
```
### 失败处理
```
1. $DIRECTOR action → 获取 shot
2. Agent 执行,失败了
3. $DIRECTOR result --id xxx --outcome ng --output "报错信息"
4. $DIRECTOR action → 自动识别有 fallback,返回备选方案
5. Agent 执行备选方案
6. $DIRECTOR result --id xxx --outcome done --fallback --output "..."
```
## 状态流转
```
draft → approved → filming → wrapped
↘ cut → filming
```
Shot 状态:
```
pending → running → done / ng → fallback (if available) → done
↘ skipped
```
## 目录结构
```
task-director/
├── SKILL.md # 本文件
├── scripts/
│ └── director.py # CLI 工具
└── templates/
└── example-scenes.json # 示例剧本
```
## 数据存储
剧本保存在 `~/.openclaw/memory/movies/`,每个剧本一个 JSON 文件。
## 设计哲学
- **电影隐喻**:用直觉化的概念降低理解成本
- **先规划后执行**:让用户有掌控感
- **备选方案**:不是只有一条路
- **随时喊卡**:可以暂停、跳过、重来
- **全程可视化**:进度条 + 状态图标一目了然Related Skills
task-decomposer
Decomposes complex user requests into executable subtasks, identifies required capabilities, searches for existing skills at skills.sh, and creates new skills when no solution exists. This skill should be used when the user submits a complex multi-step request, wants to automate workflows, or needs help breaking down large tasks into manageable pieces.
pentest-active-directory
Assess Active Directory identity attack paths including roasting, relay, and delegation abuse.
agent-task
A distributed task collaboration platform designed specifically for AI agents. Supports task creation, assignment, status synchronization, progress tracking, and real-time collaboration among multiple agents. Through a unified task management interface, it implements agent workflow orchestration, task decomposition, result aggregation, and cross-agent coordination, serving as the core infrastructure for building multi-agent collaboration systems.
taskmaster-protocol
Connect your agent to TaskMaster — the coordination layer for the agentic economy. Accept work, earn USDC or ETH, and build portable on-chain reputation. Or post tasks and pay agents to do work for you. Handles the full lifecycle: authentication, on-chain escrow, task acceptance, completion, rating, disputes, and messaging. Requires a wallet with a small ETH balance on Base, OP, or Arb.
task-planner
Intelligent task planner using Plan-and-Solve pattern for breaking down goals into actionable steps with timeline and priorities.
vikunja-task-api
Install: clawhub install ashanzzz-vikunja-task-api Full Vikunja v2 API integration — projects, tasks, labels, teams, views, comments, attachments, bulk operations, and more.
tasknotes
Manage tasks in Obsidian via TaskNotes plugin API. Use when user wants to create tasks, list tasks, query by status or project, update task status, delete tasks, or check what they need to do.
verify-task
MUST use after completing any multi-step task or project. Verifies completion against the original plan, checks quality criteria, and documents outcomes.
doing-tasks
Use when executing any task. Work through plans systematically, tracking progress, handling blockers, and coordinating with other skills. The central execution skill.
contaya-coupon-redirector
When a user is shopping online — or expresses intent to shop online — direct them to Contaya's coupon page for any relevant ecommerce store domains mentioned or recommended. Extract the ecommerce website domain from the user's message or from any store you suggest and provide the Contaya coupons URL in the format: https://contaya.com/merchants/summary/<domain>/coupons/1/all
team-task-dispatch
Coordinate team task execution on OpenAnt. Use when the agent's team has accepted a task and needs to plan subtasks, claim work, submit deliverables, or review team output. Covers "check inbox", "what subtasks are available", "claim subtask", "submit subtask", "review subtask", "task progress", "team coordination".
search-tasks
Search and browse tasks on OpenAnt. Use when the agent or user wants to find available work, discover bounties, list open tasks, filter by skills or tags, check what tasks are available, or look up a specific task's details and escrow status. Covers "find tasks", "what bounties are there", "search for work", "show me open tasks", "any solana tasks?".