CodeScene AGENTS.md Pattern: Encode Agent Decision Logic and MCP Tool Sequencing in a Single File for Predictable Refactoring Workflows

GitHub·medium signal

CodeScene's MCP Server uses an AGENTS.md file that defines the intended workflow and decision logic for AI agents — specifying how agents should sequence MCP tools during refactoring (review → plan → refactor → re-measure). This codifies the mandatory workflow including when to trigger code_health_review, how to interpret scores, and when to loop. The pattern is generalizable: any team using MCP tools can write an AGENTS.md that transforms individual tool calls into predictable, repeatable agent workflows instead of leaving sequencing to model discretion.