Why is there a 1000-line limit for expertise files?

The limit is a critical constraint to protect the LLM's context window. It ensures that the expert's mental model remains scannable, focused, and efficient without causing token bloat.

When should I avoid using Agent Expert Creation?

You should avoid building experts for simple, stable codebases or trivial tasks where the overhead of maintaining an expertise file exceeds the benefit of the automation.

Is the expertise file considered the source of truth?

No. The actual codebase is always the source of truth. The expertise file serves as the agent's mental model and working memory, which is regularly validated against the code.

What is the Act-Learn-Reuse pattern?

It is a cyclic workflow where an agent takes a useful action (Act), stores new findings in a YAML expertise file (Learn), and reads that file first during the next execution (Reuse) to maintain context.

Do I have to manually update the expertise files?

No. The system uses a 'self-improve' prompt that teaches the agent how to analyze git diffs and codebase changes to update its own expertise files automatically.

Agent Expert Creation

Name: Agent Expert Creation
Author: melodic-software

bymelodic-software

•

생산성 및 워크플로

Creates specialized AI agents that learn and maintain domain-specific mental models using the Act-Learn-Reuse pattern.

소개

Agent Expert Creation is a specialized framework for Claude Code that solves the 'agent amnesia' problem by implementing the Act-Learn-Reuse pattern. It enables developers to build domain-specific experts—such as specialists for billing, security, or database management—that maintain persistent YAML expertise files. These files act as a structured working memory or 'mental model' of your codebase, ensuring the agent retains deep context across multiple sessions. By automating the self-improvement cycle, this skill helps maintain a reliable link between the agent's knowledge and the actual state of the code, preventing repetitive mistakes in complex environments.

주요 기능

Enforces strict context window protection with 1000-line expertise limits.
Automates a Plan-Build-Improve workflow for domain-specific automation.
Maintains YAML-based expertise files as scannable mental models.
1 GitHub stars
Syncs agent knowledge with codebases via automated self-improve prompts.
Implements the Act-Learn-Reuse pattern for persistent agent memory.

사용 사례

Managing high-risk systems like payment gateways or security protocols where errors cascade.
Onboarding agents to rapidly evolving codebases with complex architectural patterns.
Creating specialized sub-agents for specific technology stacks like WebSockets, Auth, or DevOps.

소개

주요 기능

Enforces strict context window protection with 1000-line expertise limits.
Automates a Plan-Build-Improve workflow for domain-specific automation.
Maintains YAML-based expertise files as scannable mental models.
1 GitHub stars
Syncs agent knowledge with codebases via automated self-improve prompts.
Implements the Act-Learn-Reuse pattern for persistent agent memory.

사용 사례

Managing high-risk systems like payment gateways or security protocols where errors cascade.
Onboarding agents to rapidly evolving codebases with complex architectural patterns.
Creating specialized sub-agents for specific technology stacks like WebSockets, Auth, or DevOps.