self-improving-agent

What This Skill Does

The Self-Improving Agent is a critical intelligence layer for Claude Code users, designed to transform fleeting conversation logs and automated memory into durable, actionable project knowledge. By acting as a bridge between the transient nature of LLM conversations and the persistence required for long-term coding projects, this skill ensures that once a technical hurdle is cleared or a project pattern is established, it is never forgotten. It continuously parses MEMORY.md to identify recurring obstacles, architectural decisions, and successful debugging patterns. Through its suite of management commands, it allows you to curate these findings, moving them from passive logs into your active project configuration (CLAUDE.md) or modular rule files (.claude/rules/), effectively training your AI coding assistant on your specific repository standards over time.

Installation

To integrate the Self-Improving Agent into your workflow, choose the method appropriate for your environment:

• Claude Code Plugin: Execute /plugin marketplace add alirezarezvani/claude-skills followed by /plugin install self-improving-agent@claude-code-skills.
• OpenClaw CLI: Run clawhub install self-improving-agent.
• Codex CLI: Execute the install script: ./scripts/codex-install.sh --skill self-improving-agent.

Use Cases

1. Consolidating Debugging Insights: When you solve a difficult build error, use the agent to extract the solution into a reusable rule to prevent future regressions.
2. Standardizing Project Architecture: Identify inconsistent coding patterns across multiple sessions and promote the 'best' version of those patterns to your global CLAUDE.md.
3. Memory Maintenance: Keep your project knowledge base lean and performant by reviewing and pruning stale or irrelevant entries from the auto-generated MEMORY.md file.
4. Knowledge Graduation: Move complex, context-heavy notes into dedicated files in .claude/rules/ to ensure they are available to Claude the moment a related file is opened.

Example Prompts

1. "@self-improving-agent /si:review the current MEMORY.md and suggest which patterns should be promoted to CLAUDE.md."
2. "@self-improving-agent /si:extract the pattern for our custom API authentication flow into a new reusable skill rule."
3. "@self-improving-agent /si:status give me a health report on our project's memory usage and let me know if we are hitting any capacity limits."

Tips & Limitations

• Tip: Use /si:remember manually whenever you realize a specific technical decision or library quirk is something you will need to reference in the future.
• Tip: Periodically run /si:review at the end of a sprint to ensure that project-wide changes are reflected in your project rules.
• Limitation: While the agent is highly intelligent, it cannot 'delete' original logs, only suggest promotions. You should manually verify and finalize any changes made to your .claude/rules/ directory to ensure they align with current project requirements.