Claude Code vs. Cursor: Who Truly Leads Contextual Rule Management?

Claude Code has quietly added a new feature called "Modular rules". As a longtime user of Cursor, the author immediately thought of the familiar .cursor/rules system and wonders whether Claude Code is copying it.

Claude Code's new weapon: .claude/rules

The core usage is to create a .claude/rules/ directory in a project and place various .md files there to define code style, testing conventions, or security requirements. The system automatically reads all files in that directory, giving them the same priority as the root CLAUDE.md file, which provides great flexibility for teams.

Three configuration levels can be set via the paths field in the YAML front‑matter of each rule file:

Organization level : company‑wide policy.

Project level : rules for the current repository.

User level : personal preferences.

The most practical feature is the paths field. For example:

---
paths: src/api/**/*.ts
---
# API development rules
- All API endpoints must include input validation

This configuration means the rule only applies to TypeScript files under src/api/. Omitting paths makes the rule apply to all files, similar to a glob pattern.

Cursor's rule system

Cursor relies on two components: .cursor/rules and AGENTS.md. The former uses a proprietary MDC format ( .mdc) that extends Markdown with metadata such as alwaysApply. Cursor also provides a visual Settings > Rules panel for managing these files. AGENTS.md is a plain Markdown file that can be placed in the project root or any subdirectory. Cursor automatically reads it, offering a lightweight alternative for projects that do not need complex metadata.

Deep comparison: core differences

1. Trigger logic – deterministic vs. semantic

Claude Code uses deterministic path matching. A rule is applied only when the edited file matches the paths pattern, similar to a .gitignore rule.

Cursor employs semantic matching. Besides path matching, it can "Apply Intelligently" based on the rule's description, letting the AI decide whether the rule is relevant, and also supports manual invocation with @.

2. Format definition – generic vs. specialized

Claude Code sticks to pure Markdown with standard YAML front‑matter and even supports symlinks for cross‑project rule reuse.

Cursor uses the MDC format, a Markdown‑based file with additional control switches (e.g., alwaysApply) that integrates tightly with the editor UI, though it is less portable outside the editor.

3. Proximity principle

Both systems adopt a "centralized + distributed" model: a central directory ( .claude/rules or .cursor/rules) for global management, and rule files that apply locally to the nearest code.

Both support nested rule files ( CLAUDE.md for Claude Code, AGENTS.md for Cursor) that are automatically loaded when the scanner reaches a subdirectory.

Thus, the closer a rule file is to the code, the more effective it becomes for both tools.

Returning to the initial question, the author concludes that Claude Code is not merely copying Cursor; rather, both solutions have arrived at similar outcomes through independent evolution.