Agent Skills
› aiming-lab/MetaClaw
› tool-selection-strategy
tool-selection-strategy
GitHub指导智能体在Agentic工作流中高效选择工具。遵循最小化、先读后写、避免猜测调用及并行处理原则，通过决策启发式方法精准匹配工具，防止冗余操作。
Trigger Scenarios
需要决定调用哪个工具时
构建或优化Agent工作流逻辑时
Install
npx skills add aiming-lab/MetaClaw --skill tool-selection-strategy -g -y
SKILL.md
Frontmatter
{
"name": "tool-selection-strategy",
"category": "agentic",
"description": "Use this skill when deciding which tools to call in an agentic workflow. Always choose the minimal, most direct tool for each step and avoid redundant or speculative tool calls."
}
Tool Selection Strategy
Principles:
- Least tool principle: Use the most specific, lightweight tool that accomplishes the goal.
- Read before writing: Always read a file before editing it to understand current state.
- Avoid speculative calls: Don't call a tool "just to see what happens". Have a clear hypothesis.
- Parallelize independent calls: If two reads don't depend on each other, fire them simultaneously.
Decision heuristic:
- Can I answer this from memory/context? No tool call needed.
- Is this a file operation? Use Read/Write/Edit tools.
- Is this a code search? Use Grep/Glob tools.
- Is this a system operation? Use Bash (last resort).
Anti-pattern: Using a heavy tool (Agent, Bash) when a lightweight dedicated tool suffices.
Version History
- 922caf3 Current 2026-07-25 11:09
