Meta Agent Evolution

Meta Agent Evolution is an open-source system designed to make AI coding agents more reliable through continuous observation and human-supervised improvement. Integrated with OpenCode through the Model Context Protocol (MCP), it records execution events, detects recurring failure patterns across sessions, and creates evidence-backed proposals for improving agents, prompts, and skills. The system uses Node.js, JSONL, SQLite, and a dedicated WatcherRunner. Proposed changes remain pending until a human reviews them; nothing is automatically applied. The project was inspired by my earlier personal experiments, Task Observer and Skill Prompt Evolution, but was developed as a separate implementation for this hackathon without reusing their source code or Git history. The current prototype demonstrates event collection, pattern analysis, persistent pending proposals, and a supervised review workflow. Specific AI-generated code improvements remain under development.