---
title: Agentic Skill-Composition Benchmark
emoji: ⚖️
colorFrom: blue
colorTo: green
sdk: static
pinned: false
license: mit
short_description: Token cost of skill-based agents vs monolithic prompts
---

# Agentic Skill-Composition Benchmark

Measures the context-token cost and success rate of **skill-composed** agent
workflows (small skill index + on-demand loading) against a **monolithic
prompt** that inlines every skill on every turn.

Source, harness, and methodology:
https://github.com/aabhimittal/Agentic-skill-composition-benchmark