How does this help with project documentation?

It identifies recurring patterns and gaps in project knowledge, allowing you to 'promote' important findings to permanent documentation like CLAUDE.md or AGENTS.md.

Can this skill be used with other AI coding agents?

Yes, while optimized for Claude Code, the configuration and logging patterns are designed to be compatible with other agents like Pi, Codex, and OpenCode.

Does it automatically fix the errors it logs?

It captures the context, error messages, and suggested fixes which provide the agent or the developer with the necessary information to implement permanent resolutions efficiently.

What is the primary benefit of the Self-Improving Agent skill?

It creates a persistent feedback loop that allows Claude to learn from mistakes and user feedback, preventing the same errors from occurring in future sessions.

Where are the learnings and errors stored?

All data is stored locally in your project's .learnings/ directory as structured Markdown files, ensuring you have full ownership of the feedback data.

Self-Improving Agent

Name: Self-Improving Agent
Author: carmandale

bycarmandale

•

Productivity & Workflow

Captures and logs AI learnings, errors, and user corrections to enable continuous agent improvement through a structured feedback loop.

The Self-Improving Agent skill establishes a systematic way for AI coding agents to record and learn from their experiences. By creating a dedicated .learnings/ directory, the skill allows Claude to document unexpected failures, user-provided corrections, and missing capabilities in real-time. This structured data serves as a local knowledge base that the agent can reference before major tasks, ensuring that once a lesson is learned or a convention is established, it is persistently remembered and promoted to project memory.

Key Features

01Automated logging of errors, corrections, and knowledge gaps into structured Markdown.

021 GitHub stars

03Promotion workflow to elevate critical context to project memory files like CLAUDE.md.

04Integrated review triggers to scan past learnings before starting major development tasks.

05Standardized templates for tracking priority, status, and suggested resolutions.

06Recurring pattern detection through cross-referenced entry IDs and metadata.

Use Cases

01Building a project-specific knowledge base that survives across different AI sessions.

02Capturing ad-hoc feature requests and user feedback during the flow of development.

03Reducing repetitive mistakes by logging and reviewing past command or logic failures.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add carmandale/agent-config personal

For use in Claude.ai and ChatGPT

Download Skill