SkillOpt

Active
GitHub Python MIT

Description

SkillOpt is Microsoft's open-source agent skill optimization tool that selects, composes and tunes LLM tool-calling skills to improve task performance.

Key Features

  • Skill selection — auto-picks the best skill mix from a candidate pool
  • Composition — learns the optimal workflow across many skills
  • Evaluation-driven — built-in evaluator with baseline comparison
  • Explainable — surfaces the reasoning behind each pick and composition
  • Multi-model — works with GPT, Claude and local LLMs
  • Extensible — plug in custom skills and evaluators

Use Cases

💡 Evaluating and optimizing an existing agent's tool-calling strategy
💡 Searching for the best orchestration across many agent skills
💡 Turning agent tuning into a data-driven engineering practice
💡 Comparing different skill sets against standardized benchmarks

Quick Start

pip install skillopt
from skillopt import SkillOptimizer
opt = SkillOptimizer(model="gpt-4o")
opt.run(task="code-review", skills=my_skills)

Related Projects