What is a Learning Agent?

A Learning Agent is a specialized AI sub-agent that analyzes its own performance and mistakes to improve its output in future sessions.

Can I use existing agents as templates?

Yes, the create command allows you to seed new agents using the knowledge files of an existing agent as a starting point.

How do I report a mistake an agent made?

Use the report_issue command with the agent ID and specific details to provide a feedback loop for the agent's next learning cycle.

How do these agents learn?

They utilize a dedicated learning cycle that processes session logs and reported issues to update internal knowledge and refine execution strategies.

Where is the agent data stored?

Agent data, templates, and logs are stored locally within the .deepwork and .claude directories in your project repository.

Learning Agents Manager

Name: Learning Agents Manager
Author: Unsupervisedcom

byUnsupervisedcom

•

Data Science & ML

Orchestrates self-improving AI sub-agents that automatically learn from past mistakes and session transcripts to enhance future performance.

The Learning Agents skill enables Claude Code to manage a fleet of specialized sub-agents designed for continuous improvement. By analyzing session logs and user feedback, these agents execute learning cycles to identify errors and refine their strategies over time. This skill is ideal for complex, repetitive coding tasks where domain-specific knowledge evolves, allowing developers to create agents, seed them with templates, and maintain a feedback loop for persistent optimization across sessions.

Key Features

01Automated Learning Cycles: Process pending session transcripts to improve agent accuracy.

02Centralized Routing: Simple command-line interface for managing agent creation, training, and feedback.

03Context Persistence: Maintain agent knowledge and strategy improvements across multiple sessions.

0435 GitHub stars

05Issue Reporting: Capture and track specific failures to inform the learning process.

06Agent Scaffolding: Create new AI sub-agents with optional template-based seeding.

Use Cases

01Automating complex migrations where agents learn from previous edge case failures.

02Maintaining domain-specific coding standards across large, evolving repositories.

03Developing specialized agents for framework-specific tasks like background job optimization.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add unsupervisedcom/deepwork learning-agents

For use in Claude.ai and ChatGPT

Download Skill