[DOCS] Performance comparison: Subagent vs. hook vs. skill invocation overhead

Documentation Type

Missing documentation (feature not documented)

Documentation Location

https://code.claude.com/docs/en/sub-agents#create-custom-subagents

Section/Topic

https://code.claude.com/docs/zh-CN/features-overview#skill-vs-subagent

Current Documentation

_No response_

What's Wrong or Missing?

I'm trying to understand the performance overhead differences between three approaches for executing scripts/tools:

1. Subagent approach: Invoking a tool (e.g., Bash) through a subagent using the Task tool
2. Direct hook approach: Calling a script directly via a hook (e.g., PreToolUse with async: false)
3. Skill-based approach: Triggering a skill that executes a tool or script

Questions:

• What is the performance overhead of each approach compared to the others?
• Are there any latency benchmarks or performance metrics available for subagent invocation, hook execution, and skill triggering?
• How does the overhead scale with complexity (simple scripts vs. complex operations)?
• In what scenarios would one approach be significantly more efficient than the others?
• If a skill triggers an async hook, what is the combined overhead?

Context:
I'm designing a system where I need to decide between using subagents, hooks, or skills for executing scripts, and understanding the performance implications would help me make the right architectural choice.

Additional information:

• Claude Code version: e2.1.114
• OS: Linux

Suggested Improvement

The current documentation introduces Skills, Subagents, and Hooks primarily from a use-case perspective. However, to make informed architectural decisions, developers need additional guidance on:

1. Performance characteristics - execution time, latency, and throughput differences between each approach
2. Resource consumption - computational cost, token usage, and API call overhead for each option
3. Comparative analysis - a side-by-side comparison table or decision matrix showing when to use Skills vs. Subagents vs. Hooks based on performance and cost trade-offs

This would help framework users avoid blind architectural choices and select the most appropriate solution based on their specific performance and cost requirements.

Impact

High - Prevents users from using a feature

Additional Context

_No response_