Self-Evolution
Commander improves over time through a meta-learning system that tunes agent configurations based on execution outcomes. The system combines Thompson Sampling and Reflexion to optimize topology selection, provider choice, and parameter settings. Uses Thompson Sampling with Beta distributions to balance exploration and exploitation:
本文说明 Self-Evolution 在 Commander 中的职责、使用方式与相关模块。命令与代码路径与产品保持一致。
bash
Execution completes
│
├─ TrajectoryAnalyzer ← Analyze execution patterns
│ └─ Extract features: duration, tokens, success rate
│
├─ MetaLearner ← Thompson Sampling
│ └─ Update Beta distributions per arm
│要点
- 与英文源文档语义对齐;API 与 CLI 以 monorepo 为准
- 需要可运行示例时,优先使用 快速开始 中的
cliEntry.ts路径 - 指标口径:25 提供商 · 5 拓扑 · 18 工具 · 6700+ 测试