active-learn

What problem does it solve?

This Skill addresses uneven agent performance by implementing a structured, adversarial training loop that identifies weaknesses, generates targeted challenges, and reinforces learned capabilities.

Core Features & Use Cases

• Adversarial Training Loop: Runs a cycle of diagnose, challenge generation, execution, and learning updates.
• Weakness Identification: Pinpoints specific areas where an agent struggles through inline profiling and git analysis.
• Targeted Challenge Generation: Creates relevant, difficult tasks based on identified weaknesses.
• Performance Tracking: Updates learnings and capability YAML to track improvement over time.
• Use Case: When an agent repeatedly fails on specific types of code refactoring tasks, active-learn can be used to diagnose the exact nature of the failure, generate tailored refactoring challenges, and train the agent until it consistently succeeds.

Quick Start

Use the active-learn skill to train the agent named 'rust-dev' for 3 rounds.