Antigravity Customizations: Skills, Rules & Plugin Ecosystem

---
name: code-review
description: Automatically reviews Git pull request diffs for security, performance, and TypeScript type soundness
---

# Automated Code Review Protocol

When the user requests code review for the active branch or commit, execute the following:

1. **Diff Extraction**: Run `git diff HEAD~1` via `run_command` to inspect recent modifications.
2. **Security Audit**: Verify no hardcoded secrets, injection vectors, or unsanitized user inputs exist.
3. **Performance & Types**: Ensure asynchronous errors are caught and strict return types are declared.
4. **Output Format**: Format feedback as concise GitHub Flavored Markdown with exact line references.

The Three Pillars of Customization

In Antigravity 2.0, customization is not achieved by concatenating messy prompt fragments. Instead, it is partitioned across three strictly delineated systems:

+-----------------------------------------------------------+
|                    Antigravity Platform                   |
+-----------------------------------------------------------+
         |                        |                        |
         v                        v                        v
+------------------+     +------------------+     +------------------+
|   Rules Engine   |     |   Skills Engine  |     |  Plugins System  |
| - Permanent      |     | - Triggered      |     | - Bundled        |
|   constraints    |     |   on-demand      |     |   distribution   |
| - Coding styles  |     | - Isolated steps |     | - Includes MCP   |
+------------------+     +------------------+     +------------------+

---

1. Rules (Permanent Behavioral Constraints)

Rules govern the uncompromising boundaries an agent must follow at all times. They are injected directly into context at session initialization to anchor agent identity, coding styles, git branching workflows, or security guardrails.

Location Hierarchy

• Global: ~/.gemini/config/rules/*.md or ~/.gemini/config/GEMINI.md
• Workspace: .agents/rules/*.md or repository root GEMINI.md

Production Example

# Repository Engineering Rules
- All exported functions must declare explicit TypeScript return types.
- Never use console.log in production files; use src/logger.ts.
- Never create ad-hoc temporary scripts in the repo root; use scratch/.

---

2. Skills (On-Demand Workflow Playbooks)

Skills are modular, on-demand procedure playbooks. The agent only knows each skill’s name and short description during initial reasoning. Only when a user prompt matches a skill’s description does the agent read the full SKILL.md via view_file to execute its specific instructions.

Standard Skill Directory

skills/<skill_name>/
├── SKILL.md          # Core instructions with YAML Frontmatter
├── scripts/          # Helper executable scripts (Bash, Python, Node)
├── examples/         # Reference input/output patterns
└── references/       # Architecture documents & API schemas

---

3. Plugins (Team-Wide Tool Bundling)

Plugins bundle multiple related Skills, Subagents, and Sidecar MCP server definitions into a single distributable package.

Plugin Directory Layout

plugins/<plugin_name>/
├── plugin.json       # Metadata (name, version, description)
├── skills/           # Bundled skill directories
│   └── deploy-k8s/
│       └── SKILL.md
├── agents/           # Custom subagent definitions
└── mcp_config.json   # Sidecar MCP services launched with the plugin

By symlinking plugin folders into ~/.gemini/config/plugins/, engineering teams can distribute verified agentic capabilities across hundreds of developer machines in seconds.