agent-autopilot
edoserbia 作
Productivity & Tasks概要
Self-driving agent workflow with heartbeat-driven task execution, day/night progress reports, and long-term memory.
説明
このスキルでできること
A heartbeat-driven workflow that runs an autonomous agent on a ~30-minute cycle. Each cycle checks a todo list, executes the highest-priority task, and logs results. Progress reports go out during daytime hours at configurable intervals, and long-term memory is distilled from daily logs every 6 hours.
It replaces manual polling and status checks by baking reporting, memory maintenance, and task execution into a single heartbeat loop that runs without human prompting.
使いどころ
- Running a coding agent overnight without manual check-ins
- Sending structured daytime progress updates from an agent to a human manager
- Recovering an interrupted agent from its memory logs and todo state
- Keeping a quantitative strategy agent iterating through backtests autonomously
- Consolidating weeks of agent activity logs into a structured memory file
ワークフロー例
User asks: Set up agent-autopilot for a new research agent
- 1
Run init.sh with the agent workspace path to create directories and install todo-management
- 2
Write HEARTBEAT.md using the three-task template, setting report interval N to 3 hours
- 3
Create initial todo entries by breaking the project goal into concrete tasks
- 4
Agent receives heartbeat, reads todo list, sets highest-priority pending task to in_progress
- 5
Agent executes task, marks it done, appends result to memory/YYYY-MM-DD.md
Agent runs autonomously, sending daytime progress reports every 3 hours and consolidating memory every 6 hours
インストール
clawhub install edoserbia/agent-autopilotCLIコマンドを実行するか、Claude Code、Cursor、Codexなどお使いのコーディングエージェントにプロンプトを貼り付けてください。