Why Cursor AI rules silently fail and how to fix frontmatter config

Developers using Cursor's AI coding assistant often find their custom rules are ignored, not due to model errors but because of misconfigured frontmatter fields in .mdc rule files. Cursor supports four rule types — Always, Auto Attached, Agent Requested, and Manual — each activated by specific frontmatter settings like alwaysApply, globs, and description. A common mistake is formatting the globs field as a YAML array or quoted string instead of a plain comma-separated value, which silently prevents the rule from attaching. Rules saved with a .md extension instead of .mdc are also ignored without any warning. Writing a meaningful description field is equally critical for Agent Requested rules, as the AI uses that text to decide relevance — a vague title like 'Testing rules' will not trigger the rule the way a descriptive phrase will.