Platform-Specific Configurations for Claude Code, Cursor, and AI Agents in the Website Cloner Template

The JCodesMore AI Website Cloner Template maintains a single source of truth in AGENTS.md while using distinct delivery mechanisms—direct imports for Claude Code, pointer files for Cursor, and format-specific transpilation for Gemini, Amazon Q, and others—to ensure consistent instructions across diverse AI coding platforms.

The JCodesMore/ai-website-cloner-template repository solves the challenge of managing multiple AI agent configurations by centralizing core instructions and automatically adapting them to platform-native formats. Understanding these platform-specific configurations for Claude Code, Cursor, and other AI agents reveals how the template enforces consistency while respecting each tool's unique file format requirements and variable substitution syntax.

How Claude Code Reads Instructions Directly

Claude Code requires no generated copies of the master instructions. Instead, the repository provides a single CLAUDE.md file that imports AGENTS.md using the directive @AGENTS.md.

This reference-based approach means Claude Code reads the raw markdown and front-matter defined in AGENTS.md directly, eliminating the need for synchronization scripts to create secondary copies. When you update the master file, Claude Code immediately sees the changes without requiring additional build steps.

Cursor's Pointer-Based Configuration Model

Cursor takes a hybrid approach that combines pointer files with generated skill definitions. The template creates .cursor/rules/project.mdc, which tells Cursor to read AGENTS.md automatically rather than consuming a generated copy of the rules.

For skill-specific commands, the scripts/sync-skills.mjs script generates .cursor/commands/clone-website.md as a plain-markdown file. Unlike other platforms that use variable placeholders, this generated file replaces $ARGUMENTS with static descriptive text—specifically the target URL provided by the user—ensuring compatibility with Cursor's command structure while removing the need for runtime argument substitution.

Format Transpilation for CLI and Code Editor Agents

The template handles nine distinct AI agents, each requiring specific file formats and placeholder syntax. The scripts/sync-skills.mjs file contains explicit logic for each transformation:

• Codex CLI and GitHub Copilot receive the raw SKILL.md file unchanged, preserving the original markdown format and $ARGUMENTS placeholders.

• Gemini CLI receives a TOML file where $ARGUMENTS becomes {{args}}, matching Google's template syntax.

• OpenCode receives markdown with YAML front-matter, supporting native $ARGUMENTS substitution within the metadata block.

• Augment Code uses similar markdown with YAML front-matter but adds an argument-hint field for enhanced UI rendering.

• Continue requires a front-matter block containing invokable: true to mark the file as a callable skill.

• Amazon Q receives a JSON definition that includes the prompt body with $ARGUMENTS pre-replaced and an explicit list of file contexts.

The Automation Scripts Behind the Workflow

Two scripts enforce the single-source-of-truth workflow:

scripts/sync-skills.mjs parses the source skill and writes platform-specific files. The script explicitly handles Cursor's plain markdown output, Gemini's TOML conversion with {{args}} substitution, and Amazon Q's JSON structure.

scripts/sync-agent-rules.sh creates pointer files for agents that read AGENTS.md directly (like Cursor) while writing generated copies for agents requiring localized versions.

Run both scripts after editing AGENTS.md or skill definitions:

# Update all generated skill files

node scripts/sync-skills.mjs

# Regenerate pointer and rule files for agents

bash scripts/sync-agent-rules.sh

Key Configuration Files by Platform

Agent	Configuration File	Format	Source Reference
Claude Code	CLAUDE.md	Markdown import	@AGENTS.md
Cursor	.cursor/rules/project.mdc	Pointer file	References AGENTS.md directly
Cursor Skills	.cursor/commands/clone-website.md	Plain markdown	Generated from SKILL.md
Gemini CLI	.gemini/commands/clone-website.toml	TOML	Generated with {{args}}
Amazon Q	.amazonq/cli-agents/clone-website.json	JSON	Pre-substituted arguments
OpenCode	.opencode/commands/clone-website.md	Markdown + YAML	Front-matter with arguments
Augment Code	.augment/commands/clone-website.md	Markdown + YAML	Includes argument-hint
Continue	.continue/commands/clone-website.md	Markdown + YAML	invokable: true flag
GitHub Copilot	.github/skills/clone-website/SKILL.md	Raw markdown	Unmodified SKILL.md

Summary

• Claude Code uniquely imports AGENTS.md directly via CLAUDE.md using the @AGENTS.md directive, requiring no generated copies.
• Cursor uses a pointer file (.cursor/rules/project.mdc) to reference AGENTS.md and receives plain-markdown skill files without variable placeholders.
• Nine distinct formats are supported through automated transpilation, including TOML for Gemini, JSON for Amazon Q, and YAML front-matter for OpenCode and Augment.
• sync-skills.mjs and sync-agent-rules.sh maintain synchronization after any changes to the master instruction files.
• All agents ultimately derive from the same source content, ensuring consistency while respecting platform-specific capabilities.

Frequently Asked Questions

Why does Claude Code use @AGENTS.md while other agents require generated files?

Claude Code supports direct file referencing within its instruction files, allowing it to parse AGENTS.md dynamically at runtime. Other agents like Cursor, Gemini, and Amazon Q expect their configuration files to exist in specific locations with platform-native syntax, necessitating the generation scripts to create compatible copies or pointers.

What is the difference between AGENTS.md and SKILL.md in this template?

AGENTS.md serves as the master instruction set defining general agent behavior and code review guidelines applicable to all AI agents. SKILL.md (located in .claude/skills/clone-website/) contains the specific command definition for the website cloning functionality, which gets transpiled into various formats (TOML, JSON, etc.) depending on the target agent's requirements.

How do variable placeholders differ between platforms?

The template handles three distinct substitution patterns: $ARGUMENTS works natively in OpenCode and the raw SKILL.md, transforms to {{args}} for Gemini's TOML format, and gets pre-replaced with static descriptions (like the target URL provided by the user) for Cursor and Amazon Q's JSON output.

Which agents require running the sync scripts after updating instructions?

You must run sync-agent-rules.sh whenever you modify AGENTS.md to update pointer files and generated rule copies. Run sync-skills.mjs whenever you modify the SKILL.md source to regenerate platform-specific skill files. Claude Code requires neither script for AGENTS.md changes due to its direct import capability, though CLAUDE.md itself should be committed to the repository.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}