ReflexioAI/claude-smart

What it solves

Claude-smart is a self-improvement plugin for AI coding assistants (Claude Code, Codex, and OpenCode) that prevents the assistant from repeating the same mistakes. Instead of just recording a history of what happened, it converts user corrections and successful workflows into durable, reusable rules (skills) that the AI follows in future sessions.

How it works

The plugin uses lifecycle hooks to capture every user turn, tool call, and assistant response. These interactions are processed by the Reflexio engine to extract three types of artifacts:

• Preferences: Project-scoped settings regarding your stack, role, and communication style.
• Project-specific skills: Rules learned from corrections within a specific repository (e.g., "always use pnpm dev:all for this repo").
• Shared skills: General patterns that transfer across different projects.

On each new prompt, the system performs a hybrid search (vector + BM25) using a local ONNX embedder to find and inject only the relevant learned rules into the AI's context. All data is stored locally on the user's machine.

Who it’s for

Developers using Claude Code, Codex, or OpenCode who want their AI assistant to remember project-specific quirks, avoid repeating errors, and reduce token usage by providing distilled context instead of long conversation histories.

Highlights

• Automatic Learning: Extracts rules from interactions without manual input, though users can explicitly flag corrections using /claude-smart:learn.
• Self-Tuning: Continuously refines, merges, and archives rules as new evidence accumulates to prevent prompt bloat.
• Local-First Privacy: Runs semantic search and data storage entirely on the machine using an in-process ONNX embedder, requiring no external API calls for retrieval.
• Management Dashboard: Includes a web UI for browsing session histories and editing learned preferences and skills.