herobrine

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Scheduled Autonomous Agents

可调度的自主Agents

Run Claude Code on autopilot. Create agents that execute any prompt on a schedule — morning briefings, overnight builds, recurring research — delivered straight to your Beeper.

Claude Code isn't just interactive. It's a daemon.

让Claude Code自动运行。创建可按计划执行任意提示词的Agents——早间简报、夜间构建、周期性研究——结果直接发送到你的Beeper。

Claude Code不只是交互式工具，它还是一个守护进程。

HOW IT WORKS

工作原理

┌─────────────────┐     ┌──────────────┐     ┌─────────────────┐
│  /herobrine │────▶│  macOS       │────▶│  claude -p      │
│  create          │     │  launchd     │     │  (headless)     │
└─────────────────┘     └──────────────┘     └────────┬────────┘
                                                       │
                                              ┌────────▼────────┐
                                              │  Deliver via     │
                                              │  Beeper + notif  │
                                              └─────────────────┘

You define an agent: a name, a prompt, and a schedule
A macOS launchd plist is created to trigger it on time
The runner executes
```
claude -p
```
with
```
--dangerously-skip-permissions
```
(fully autonomous)
Results are sent to your Beeper chat and saved to local logs
macOS notification confirms completion

┌─────────────────┐     ┌──────────────┐     ┌─────────────────┐
│  /herobrine │────▶│  macOS       │────▶│  claude -p      │
│  create          │     │  launchd     │     │  (headless)     │
└─────────────────┘     └──────────────┘     └────────┬────────┘
                                                       │
                                              ┌────────▼────────┐
                                              │  Deliver via     │
                                              │  Beeper + notif  │
                                              └─────────────────┘

你定义一个Agent：名称、提示词和调度计划
创建一个macOS launchd plist文件来定时触发它
运行器使用
```
--dangerously-skip-permissions
```
参数执行
```
claude -p
```
（完全自主运行）
结果发送到你的Beeper聊天窗口并保存到本地日志
macOS通知会确认任务完成

SETUP

设置步骤

1. Install the scripts

1. 安装脚本

Copy the runner and manager scripts to

~/.claude/agents/

bash

mkdir -p ~/.claude/agents/schedules ~/.claude/agents/logs
cp scripts/run-agent.sh ~/.claude/agents/run-agent.sh
cp scripts/manage-agent.sh ~/.claude/agents/manage-agent.sh
chmod +x ~/.claude/agents/run-agent.sh ~/.claude/agents/manage-agent.sh

将运行器和管理脚本复制到

~/.claude/agents/

bash

mkdir -p ~/.claude/agents/schedules ~/.claude/agents/logs
cp scripts/run-agent.sh ~/.claude/agents/run-agent.sh
cp scripts/manage-agent.sh ~/.claude/agents/manage-agent.sh
chmod +x ~/.claude/agents/run-agent.sh ~/.claude/agents/manage-agent.sh

2. Configure Beeper delivery (optional)

2. 配置Beeper交付（可选）

Find your Beeper "Note to Self" chat ID and update the default

DELIVERY_CHAT

manage-agent.sh

. The agent will send full results as a formatted Beeper message after each run.

If you don't use Beeper, results are still saved to

~/.claude/agents/logs/

and delivered via macOS notification.

找到你的Beeper“给自己的笔记”聊天ID，更新

manage-agent.sh

中的默认

DELIVERY_CHAT

值。每次运行后，Agent会将完整结果以格式化的Beeper消息发送给你。

如果你不使用Beeper，结果仍会保存到

~/.claude/agents/logs/

目录，并通过macOS通知交付。

ACTIONS

操作说明

Determine the action

确定操作类型

$ARGUMENTS

contains an action keyword (create, list, remove, run-now, logs), use that. Otherwise, ask the user with AskUserQuestion:

Create a new scheduled agent
List all scheduled agents
Remove a scheduled agent
Run now — trigger an agent immediately
View logs of past agent runs

如果

$ARGUMENTS

中包含操作关键词（create、list、remove、run-now、logs），则执行对应操作。否则，通过AskUserQuestion询问用户：

创建新的可调度Agent
列出所有可调度Agent
删除某个可调度Agent
立即运行——立即触发一个Agent
查看日志——查看Agent过往运行的日志

Action: Create

操作：创建

Gather from the user (via AskUserQuestion or arguments):

Agent name — short identifier, lowercase with hyphens (e.g.,
```
stock-news
```
,
```
nightly-review
```
)
Prompt — the full prompt sent to Claude. Be detailed. Can include skill invocations like
```
/steve
```
or
```
/gsd:execute-plan
```
Schedule — friendly options:
- Every morning at 8am →
```
0 8 * * *
```
- Every evening at 6pm →
```
0 18 * * *
```
- Every hour →
```
every-1h
```
- Every 30 minutes →
```
every-30m
```
- Weekday mornings at 9am →
```
0 9 * * 1-5
```
- Custom cron or interval
Model —
```
opus
```
,
```
sonnet
```
(default), or
```
haiku
```
Max turns — default:
```
50
```

Then run:

bash

bash ~/.claude/agents/manage-agent.sh create "<name>" "<prompt>" "<schedule>" "$HOME" "<model>" <max_turns>

通过AskUserQuestion或参数从用户处收集以下信息：

Agent名称——简短标识符，小写并使用连字符（例如：
```
stock-news
```
、
```
nightly-review
```
）
提示词——发送给Claude的完整提示词，越详细越好。可包含技能调用，如
```
/steve
```
或
```
/gsd:execute-plan
```
调度计划——友好选项：
- 每天早上8点 →
```
0 8 * * *
```
- 每天晚上6点 →
```
0 18 * * *
```
- 每小时 →
```
every-1h
```
- 每30分钟 →
```
every-30m
```
- 工作日早上9点 →
```
0 9 * * 1-5
```
- 自定义cron表达式或时间间隔
模型——
```
opus
```
、
```
sonnet
```
（默认）或
```
haiku
```
最大轮次——默认值：
```
50
```

然后运行：

bash

bash ~/.claude/agents/manage-agent.sh create "<name>" "<prompt>" "<schedule>" "$HOME" "<model>" <max_turns>

Action: List

操作：列出

bash

bash ~/.claude/agents/manage-agent.sh list

bash

bash ~/.claude/agents/manage-agent.sh list

Action: Remove

操作：删除

If no agent name given, list first, then ask which to remove.

bash

bash ~/.claude/agents/manage-agent.sh remove "<name>"

如果未指定Agent名称，先列出所有Agent，再询问要删除哪一个。

bash

bash ~/.claude/agents/manage-agent.sh remove "<name>"

Action: Run Now

操作：立即运行

bash

bash ~/.claude/agents/manage-agent.sh run-now "<name>"

bash

bash ~/.claude/agents/manage-agent.sh run-now "<name>"

Action: Logs

操作：查看日志

bash

bash ~/.claude/agents/manage-agent.sh logs [agent-name]

bash

bash ~/.claude/agents/manage-agent.sh logs [agent-name]

USE CASES

使用场景

Agent	Prompt	Schedule
`stock-news`	Research today's top stock market movements and tech earnings. Summarize the 5 most important things.	`0 8 * * 1-5`
`pr-digest`	Review all open PRs in my-org/my-repo. Summarize changes, flag concerns, note approvals needed.	`0 9 * * 1-5`
`overnight-build`	Use /steve to implement the next phase in /path/to/project	`0 2 * * *`
`security-scan`	Audit /path/to/repo for OWASP top 10 vulnerabilities. Report findings with severity.	`0 6 * * 1`
`dep-updater`	Check for outdated dependencies in /path/to/project. Create a PR with safe updates.	`0 3 * * 0`

Agent	提示词	调度计划
`stock-news`	研究今日股市主要动向和科技公司财报。总结5个最重要的信息点。	`0 8 * * 1-5`
`pr-digest`	查看my-org/my-repo中的所有未合并PR。总结变更内容、标记问题、记录需要审批的项。	`0 9 * * 1-5`
`overnight-build`	使用/steve实现/path/to/project中的下一阶段任务	`0 2 * * *`
`security-scan`	审计/path/to/repo中的OWASP十大漏洞。按严重程度报告发现的问题。	`0 6 * * 1`
`dep-updater`	检查/path/to/project中的过时依赖项。创建包含安全更新的PR。	`0 3 * * 0`

FILE STRUCTURE

文件结构

~/.claude/agents/
├── run-agent.sh              # Executes claude -p, delivers results
├── manage-agent.sh           # Create/list/remove/run-now/logs
├── schedules/                # Agent configs (JSON)
│   └── stock-news.json
└── logs/                     # Execution logs
    ├── stock-news_20260211_080000.log
    └── stock-news_launchd.log

~/Library/LaunchAgents/
└── com.claude.agent.stock-news.plist   # macOS scheduler

~/.claude/agents/
├── run-agent.sh              # 执行claude -p，交付结果
├── manage-agent.sh           # 创建/列出/删除/立即运行/查看日志
├── schedules/                # Agent配置文件（JSON格式）
│   └── stock-news.json
└── logs/                     # 执行日志
    ├── stock-news_20260211_080000.log
    └── stock-news_launchd.log

~/Library/LaunchAgents/
└── com.claude.agent.stock-news.plist   # macOS调度器文件

KEY DETAILS

关键细节

Agents run with
```
--dangerously-skip-permissions
```
— fully autonomous, no prompts
Each agent gets its own launchd plist for reliable scheduling
The Beeper delivery instruction is injected into the prompt, so the spawned Claude instance handles delivery itself via MCP
Logs capture full output for every run with timestamps
Agents can invoke other skills (
```
/steve
```
,
```
/rnv
```
,
```
/gsd:execute-plan
```
) — it's a full Claude Code instance

Agents使用
```
--dangerously-skip-permissions
```
参数运行——完全自主，无需人工确认
每个Agent都有独立的launchd plist文件，确保调度可靠性
Beeper交付指令会被注入到提示词中，因此启动的Claude实例会通过MCP自行处理交付
日志会记录每次运行的完整输出并带有时间戳
Agents可以调用其他技能（
```
/steve
```
、
```
/rnv
```
、
```
/gsd:execute-plan
```
）——它是一个完整的Claude Code实例