Agent Skills
› aiming-lab/MetaClaw
› plan-before-multi-step-execution
plan-before-multi-step-execution
GitHub指导 Agent 在执行多步或不可逆任务前先制定详细计划，分解步骤、识别依赖与风险，验证后再执行，避免盲目操作。
触发场景
需要执行3个以上步骤的复杂任务
涉及不可逆操作的任务
步骤间存在强依赖关系的任务
安装
npx skills add aiming-lab/MetaClaw --skill plan-before-multi-step-execution -g -y
SKILL.md
Frontmatter
{
"name": "plan-before-multi-step-execution",
"category": "agentic",
"description": "Use this skill before executing a sequence of 3 or more steps, especially when steps are irreversible or depend on each other. Write out the plan and verify it before starting execution."
}
Plan Before Multi-Step Execution
For complex tasks, plan first and execute second.
Planning phase:
- Decompose the task into concrete, ordered steps.
- Identify dependencies between steps.
- Flag irreversible actions that need user confirmation.
- Identify what can fail and what the recovery path is.
Execution phase:
- Follow the plan step by step; update it if you discover new information.
- After each step, verify the expected output before proceeding.
- If a step fails, re-evaluate the remaining plan — don't blindly continue.
Anti-pattern: Starting to execute before understanding the full scope of the task.
版本历史
- 922caf3 当前 2026-07-25 11:08
