experience-library

What problem does it solve?

This Skill enables AI agents to continually learn and improve from past task outcomes without the prohibitive cost and time of model fine-tuning. It captures successful patterns and applies them as "token priors" to new tasks, accelerating learning and optimizing tool usage.

Core Features & Use Cases

• Continual Learning: Automatically scores task outcomes and extracts reusable patterns from successful executions.
• Cost-Effective Improvement: Achieves continual learning for a fraction of the cost ($18/100 samples vs. $10k fine-tune), making AI adaptation accessible.
• Flexible Pattern Application: Stores patterns as token context, allowing for dynamic application to new tasks without altering model weights.
• Use Case: After successfully extracting data from a JSON API, this skill captures the effective approach (e.g., "use requests with retry logic"). The next time you need to "Fetch user data from REST API," the AI automatically leverages this learned pattern, leading to faster and more reliable execution.

Quick Start

Example: Capture and learn from a task's experience

def update_experience_library(task, answer, outcome): score = evaluate_outcome(answer, outcome) if score > threshold: pattern = extract_pattern(task, answer) library.add_rule(pattern) return library.get_relevant_rules(new_task)

task = "Extract data from JSON API" answer = execute_task(task) # Assume this runs the task outcome = {"success": True, "time": 5.2} update_experience_library(task, answer, outcome)