Claugravity: One-Click Migration of Claude Code Plugins to Antigravity

Migration Pain Point: When Claude Code Meets Antigravity

If you use both Claude Code and Google Antigravity, you may run into an awkward problem: the extension formats of the two are completely incompatible. Claude uses CLAUDE.md files and slash commands, while Antigravity requires SKILL.md folders, YAML frontmatter, rule mappings in AGENTS.md, and configuration files under ~/.gemini/. According to feedback from the developer community, this format gap makes it difficult for users to reuse existing plugin assets between the two.

The open-source tool Claugravity was created precisely to solve this pain point. It can read existing Claude plugin repositories or folders, convert the instruction structure, and deploy assets directly into your native Antigravity environment.

Claude Code is a command-line AI coding agent launched by Anthropic, whose extension mechanism centers on the CLAUDE.md file in the project root—developers can write project conventions, coding standards, and contextual prompts there, and Claude Code automatically reads it during a session. Slash commands are shortcut instructions defined as Markdown files (such as /review and /test), stored in a specific directory for users to trigger manually. The advantage of this mechanism is that it is purely file-based with zero dependencies, but it also means that extensibility is confined to Anthropic's own ecosystem.

Google Antigravity is an AI coding platform launched by Google, covering both desktop application and IDE plugin forms. It adopts an extension model called Skill: each skill is an independent folder containing SKILL.md as the entry-point documentation, and declares metadata such as the skill name, trigger conditions, and permission scope through YAML frontmatter. Global rules are written centrally in AGENTS.md, and runtime configuration is stored under ~/.gemini/ in the user's home directory. This structure is stricter than Claude's single-file approach and facilitates permission isolation and skill discovery on the platform, at the cost of not being directly interoperable with Claude's assets.

YAML frontmatter refers to the metadata block at the beginning of a Markdown file, wrapped by two lines of ---, written in YAML syntax as key-value pairs. In Antigravity's SKILL.md, it serves as declarative configuration, with typical fields including the skill identifier, description, allowed tools to invoke, and applicable scenarios. When loading a skill, the platform first parses this metadata and then decides whether to include the skill in the available list and whether user authorization is required. Therefore, if the frontmatter is written incorrectly or is missing, the skill is often silently ignored—which is also one of the most error-prone steps in format migration.

How Claugravity Works

Claugravity is a locally running open-source tool under the MIT license, with its code hosted on GitHub. Its core function is to parse the CLAUDE.md and slash commands in Claude plugins, convert them into the SKILL.md folder structure required by Antigravity, and automatically generate YAML frontmatter and AGENTS.md rule mappings. The converted assets are placed in the ~/.gemini/ configuration directory for the Antigravity application or IDE to read.

The entire process is completed locally, with no need to upload plugin code to the cloud—a plus for privacy-conscious developers.

Featured Capabilities: Security Scanning and Conflict Handling

Beyond conversion, Claugravity also offers two noteworthy capabilities.

Static Security Scanning

Before activating any skill, the tool runs static checks, with inspection rules referencing the NVIDIA Agent Skills guidelines. It can flag risky shell patterns (such as curl commands executed through a pipe) as well as API credentials exposed in prompts. According to the project documentation, this design is intended to prevent credential leaks caused by malice or negligence, and is especially suitable when sharing plugins across a team.

Static security scanning refers to analyzing text content through pattern matching and rule sets to discover potential risks without executing code. In AI skill scenarios, the scanning targets are mainly skill description files and the command snippets within them: on one hand, it identifies dangerous shell patterns, such as piping remote content directly to the shell for execution (curl ... | bash), recursive deletion, and privilege escalation commands; on the other hand, it uses regular expressions or entropy detection to identify hardcoded API keys, tokens, and other credentials. The limitation of this type of check is that it can only cover known patterns and cannot defend against obfuscated or logically hidden malicious behavior, so it is usually the first line of defense before sharing plugins rather than the only safeguard.

Conflict Handling

When duplicate skills exist in the built-in directory and external directories, Claugravity identifies them so you can clean up the conflicts and avoid loading the wrong version at runtime.

Usage: CLI and Web Dashboard

Claugravity provides two management interfaces: a terminal CLI and a local Web dashboard. The Web dashboard runs on localhost:4321, making it convenient to visually inspect converted skills and configurations. In terms of compatibility, the tool supports Windows, macOS, and Linux, covering both the Antigravity desktop application and IDE forms.

One detail to note: Antigravity only reads configuration at startup, so after installing or updating a plugin package, you must restart the application for it to take effect.

Local Web dashboard refers to a lightweight HTTP service started by the tool itself on the development machine (here listening on localhost:4321), which users access through a browser to view and operate on data. Compared with the CLI, it is better suited to displaying structured information such as skill lists, conversion logs, and conflict comparisons. Because the service binds only to the loopback address, it cannot be accessed from external networks, and data does not leave the local machine—consistent with the tool's overall localized, privacy-first positioning. When using it, watch out for port occupation and change the port in the configuration if necessary.

Summary

Claugravity solves a specific ecosystem fragmentation problem: the extension formats of Claude Code and Antigravity are not interoperable. Through automatic conversion, security scanning, and conflict management, it reduces the cost of migrating plugins between the two AI coding environments. For developers who use both tools, this is a local tool worth trying. The project is under the MIT license and is available on GitHub.

One-line Summary

Claugravity is an open-source tool that automatically converts Claude Code plugins into the Antigravity skill format, with built-in security scanning.

Paragraph Summary

Claugravity is a locally running, MIT-licensed open-source tool designed to solve the incompatibility between the extension formats of Claude Code and Google Antigravity. It parses the CLAUDE.md and slash commands in Claude plugins, converts them into the SKILL.md folder structure required by Antigravity, automatically generates YAML frontmatter and AGENTS.md rule mappings, and deploys them to the ~/.gemini/ directory for the application or IDE to read. The entire conversion is completed locally without uploading to the cloud. The tool also includes static security scanning, referencing the NVIDIA Agent Skills guidelines, which can flag dangerous shell pipe patterns and API credentials exposed in prompts, while also identifying duplicate skills to handle conflicts. For use, it provides a CLI and a Web dashboard on localhost:4321, supports Windows, macOS, and Linux, and covers both desktop application and IDE forms; after installation or updates, Antigravity must be restarted for changes to take effect.