How to Use Antigravity Multi-Agent Skills

If you are not sure where to start, type /coordinate <your task prompt> first.

Quick Start

1. Open in Antigravity IDE

   antigravity open /path/to/oh-my-ag
   

2. Skills are automatically detected. Antigravity scans .agent/skills/ and indexes all available skills.

3. Chat in the IDE. Describe what you want to build.

Usage Examples

Example 1: Simple Single-Domain Task

You type:

"Create a login form component with email and password fields using Tailwind CSS"

What happens:

• Antigravity detects this matches frontend-agent
• The skill loads automatically (Progressive Disclosure)
• You get a React component with TypeScript, Tailwind, form validation

Example 2: Complex Multi-Domain Project

You type:

"Build a TODO app with user authentication"

What happens:

1. Workflow Guide activates — detects multi-domain complexity
2. PM Agent plans — creates task breakdown with priorities
3. You spawn agents via CLI:

   oh-my-ag agent:spawn backend "JWT authentication API" session-01 &
   oh-my-ag agent:spawn frontend "Login and TODO UI" session-01 &
   wait
   

4. Agents work in parallel — save outputs to Knowledge Base
5. You coordinate — review .agent/brain/ for consistency
6. QA Agent reviews — security/performance audit
7. Fix & iterate — re-spawn agents with corrections

Example 3: Bug Fixing

You type:

"There's a bug — clicking login shows 'Cannot read property map of undefined'"

What happens:

1. debug-agent activates — analyzes error
2. Root cause found — component maps over todos before data loads
3. Fix provided — loading states and null checks added
4. Regression test written — ensures bug won't return
5. Similar patterns found — proactively fixes 3 other components

Example 4: CLI-based Parallel Execution

# Single agent (workspace auto-detected)
oh-my-ag agent:spawn backend "Implement JWT auth API" session-01

# Parallel agents
oh-my-ag agent:spawn backend "Implement auth API" session-01 &
oh-my-ag agent:spawn frontend "Create login form" session-01 &
oh-my-ag agent:spawn mobile "Build auth screens" session-01 &
wait

Monitor in real-time:

# Terminal (separate terminal window)
bunx oh-my-ag dashboard

# Or browser
bunx oh-my-ag dashboard:web
# → http://localhost:9847

Real-time Dashboards

Terminal Dashboard

bunx oh-my-ag dashboard

Watches .serena/memories/ using fswatch (macOS) or inotifywait (Linux). Displays a live table with session status, agent states, turns, and latest activity. Updates automatically when memory files change.

Requirements:

• macOS: brew install fswatch
• Linux: apt install inotify-tools

Web Dashboard

bun install          # first time only
bunx oh-my-ag dashboard:web

Open http://localhost:9847 in your browser. Features:

• Real-time updates via WebSocket (event-driven, not polling)
• Auto-reconnect if the connection drops
• Serena-themed UI with purple accent colors
• Session status — ID and running/completed/failed state
• Agent table — name, status (with colored dots), turn count, task description
• Activity log — latest changes from progress and result files

The server watches .serena/memories/ using chokidar with debounce (100ms). Only changed files trigger reads — no full re-scan.

Key Concepts

Progressive Disclosure

Antigravity automatically matches requests to skills. You never manually select a skill. Only the needed skill loads into context.

Token-Optimized Skill Design

Each skill uses a two-layer architecture for maximum token efficiency:

• SKILL.md (~40 lines): Identity, routing, core rules — loaded immediately
• resources/: Execution protocols, examples, checklists, error playbooks — loaded on-demand

Shared resources live in _shared/ (not a skill) and are referenced by all agents:

• Chain-of-thought execution protocols with 4-step workflow
• Few-shot input/output examples for mid-tier model guidance
• Error recovery playbooks with "3 strikes" escalation
• Reasoning templates for structured multi-step analysis
• Context budget management for Flash/Pro model tiers
• Automated verification via verify.sh
• Cross-session lessons learned accumulation

CLI Agent Spawning

Use oh-my-ag agent:spawn to run agents via CLI. Respects agent_cli_mapping in user-preferences.yaml to select the appropriate CLI (gemini, claude, codex, qwen) per agent type. Workspace is auto-detected from common monorepo conventions, or can be set explicitly with -w.

Knowledge Base

Agent outputs stored at .agent/brain/. Contains plans, code, reports, and coordination notes.

Serena Memory

Structured runtime state at .serena/memories/. The orchestrator writes session info, task boards, per-agent progress, and results. Dashboards watch these files for monitoring.

Workspaces

Agents can work in separate directories to avoid conflicts. Workspace is auto-detected from common monorepo conventions:

./apps/api   or ./backend   → Backend Agent workspace
./apps/web   or ./frontend  → Frontend Agent workspace
./apps/mobile or ./mobile   → Mobile Agent workspace

Available Skills

Workflow Commands

Type these in Antigravity IDE chat to trigger step-by-step workflows:

These are separate from skills (which auto-activate). Workflows give you explicit control over multi-step processes.

Typical Workflows

Workflow A: Single Skill

You: "Create a button component"
  → Antigravity loads frontend-agent
  → Get component immediately

Workflow B: Multi-Agent Project (Auto)

You: "Build a TODO app with authentication"
  → workflow-guide activates automatically
  → PM Agent creates plan
  → You spawn agents via CLI (oh-my-ag agent:spawn)
  → Agents work in parallel
  → QA Agent reviews
  → Fix issues, iterate

Workflow B-2: Multi-Agent Project (Explicit)

You: /coordinate
  → Step-by-step guided workflow
  → PM planning → plan review → agent spawning → monitoring → QA review

Workflow C: Bug Fixing

You: "Login button throws TypeError"
  → debug-agent activates
  → Root cause analysis
  → Fix + regression test
  → Similar patterns checked

Workflow D: CLI Orchestration with Dashboard

Terminal 1: bunx oh-my-ag dashboard:web
Terminal 2: oh-my-ag agent:spawn backend "task" session-01 &
            oh-my-ag agent:spawn frontend "task" session-01 &
Browser:    http://localhost:9847 → real-time status

Tips

1. Be specific — "Build a TODO app with JWT auth, React frontend, FastAPI backend" is better than "make an app"
2. Use CLI spawning for multi-domain projects — don't try to do everything in one chat
3. Review Knowledge Base — check .agent/brain/ for API consistency
4. Iterate with re-spawns — refine instructions, don't start over
5. Use dashboards — bunx oh-my-ag dashboard or bunx oh-my-ag dashboard:web to monitor orchestrator sessions
6. Separate workspaces — assign each agent its own directory

Troubleshooting

CLI Commands

bunx oh-my-ag                # Interactive skill installer
bunx oh-my-ag doctor         # Check setup & repair missing skills
bunx oh-my-ag doctor --json  # JSON output for CI/CD
bunx oh-my-ag update         # Update skills to latest version
bunx oh-my-ag stats          # View productivity metrics
bunx oh-my-ag stats --reset  # Reset metrics
bunx oh-my-ag retro          # Session retrospective (learnings & next steps)
bunx oh-my-ag dashboard      # Terminal real-time dashboard
bunx oh-my-ag dashboard:web  # Web dashboard (http://localhost:9847)
bunx oh-my-ag help           # Show help

For Developers (Integration Guide)

If you want to integrate these skills into your existing Antigravity project, see Existing Project Integration for:

• Quick 3-step integration
• Full dashboard integration
• Customizing skills for your tech stack
• Troubleshooting and best practices

Just chat in Antigravity IDE. For monitoring, use the dashboards. For CLI execution, use the orchestrator scripts. To integrate into your existing project, use Existing Project Integration.