MCP Server: io.github.ri7in/rule-drift

The io.github.ri7in/rule-drift MCP server tests whether AI agents retain critical instructions as conversations grow longer or become distracting. It is intended to be run locally as part of a test suite when changing a prompt, model, memory system, or agent workflow.

🛠️ Key Features

• Tests rule retention under long or distracting conversation conditions
• Runs locally and does not sit in front of production requests
• Uses Node.js 22+ and Ollama

🚀 Use Cases

• Verify instruction-following after prompt changes
• Regression-test agent behavior after updating model or memory system
• Check agent workflow changes for instruction drift

⚡ Developer Benefits

• Supports test-driven validation of agent prompts and workflows
• Helps detect critical instruction drift during development

⚠️ Limitations

• Requires Node.js 22+ and Ollama to run

Known Topics

• agent-evaluation, ai-agents, evals, llm, ollama, testing, typescript, mcp, mcp-server, prompt-testing, regression-testing