melikhanmutlu/parallel-agents

Native Parallel Agents

Orchestration through Antigravity's built-in Agent Tool

Overview

This skill enables coordinating multiple specialized agents through Antigravity's native agent system. Unlike external scripts, this approach keeps all orchestration within Antigravity's control.

When to Use Orchestration

✅ Good for:

• Complex tasks requiring multiple expertise domains
• Code analysis from security, performance, and quality perspectives
• Comprehensive reviews (architecture + security + testing)
• Feature implementation needing backend + frontend + database work

❌ Not for:

• Simple, single-domain tasks
• Quick fixes or small changes
• Tasks where one agent suffices

---

Native Agent Invocation

Single Agent

Use the security-auditor agent to review authentication

Sequential Chain

First, use the explorer-agent to discover project structure.
Then, use the backend-specialist to review API endpoints.
Finally, use the test-engineer to identify test gaps.

With Context Passing

Use the frontend-specialist to analyze React components.
Based on those findings, have the test-engineer generate component tests.

Resume Previous Work

Resume agent [agentId] and continue with additional requirements.

---

Orchestration Patterns

Pattern 1: Comprehensive Analysis

Agents: explorer-agent → [domain-agents] → synthesis

1. explorer-agent: Map codebase structure
2. security-auditor: Security posture
3. backend-specialist: API quality
4. frontend-specialist: UI/UX patterns
5. test-engineer: Test coverage
6. Synthesize all findings

Pattern 2: Feature Review

Agents: affected-domain-agents → test-engineer

1. Identify affected domains (backend? frontend? both?)
2. Invoke relevant domain agents
3. test-engineer verifies changes
4. Synthesize recommendations

Pattern 3: Security Audit

Agents: security-auditor → penetration-tester → synthesis

1. security-auditor: Configuration and code review
2. penetration-tester: Active vulnerability testing
3. Synthesize with prioritized remediation

---

Available Agents

| Agent | Expertise | Trigger Phrases | |-------|-----------|-----------------| | orchestrator | Coordination | "comprehensive", "multi-perspective" | | security-auditor | Security | "security", "auth", "vulnerabilities" | | penetration-tester | Security Testing | "pentest", "red team", "exploit" | | backend-specialist | Backend | "API", "server", "Node.js", "Express" | | frontend-specialist | Frontend | "React", "UI", "components", "Next.js" | | test-engineer | Testing | "tests", "coverage", "TDD" | | devops-engineer | DevOps | "deploy", "CI/CD", "infrastructure" | | database-architect | Database | "schema", "Prisma", "migrations" | | mobile-developer | Mobile | "React Native", "Flutter", "mobile" | | api-designer | API Design | "REST", "GraphQL", "OpenAPI" | | debugger | Debugging | "bug", "error", "not working" | | explorer-agent | Discovery | "explore", "map", "structure" | | documentation-writer | Documentation | "write docs", "create README", "generate API docs" | | performance-optimizer | Performance | "slow", "optimize", "profiling" | | project-planner | Planning | "plan", "roadmap", "milestones" | | seo-specialist | SEO | "SEO", "meta tags", "search ranking" | | game-developer | Game Development | "game", "Unity", "Godot", "Phaser" |

---

Antigravity Built-in Agents

These work alongside custom agents:

| Agent | Model | Purpose | |-------|-------|---------| | Explore | Haiku | Fast read-only codebase search | | Plan | Sonnet | Research during plan mode | | General-purpose | Sonnet | Complex multi-step modifications |

Use Explore for quick searches, custom agents for domain expertise.

---

Synthesis Protocol

After all agents complete, synthesize:

## Orchestration Synthesis

### Task Summary
[What was accomplished]

### Agent Contributions
| Agent | Finding |
|-------|---------|
| security-auditor | Found X |
| backend-specialist | Identified Y |

### Consolidated Recommendations
1. **Critical**: [Issue from Agent A]
2. **Important**: [Issue from Agent B]
3. **Nice-to-have**: [Enhancement from Agent C]

### Action Items
- [ ] Fix critical security issue
- [ ] Refactor API endpoint
- [ ] Add missing tests

---

Best Practices

1. Available agents - 17 specialized agents can be orchestrated
2. Logical order - Discovery → Analysis → Implementation → Testing
3. Share context - Pass relevant findings to subsequent agents
4. Single synthesis - One unified report, not separate outputs
5. Verify changes - Always include test-engineer for code modifications

---

Key Benefits

• ✅ Single session - All agents share context
• ✅ AI-controlled - Claude orchestrates autonomously
• ✅ Native integration - Works with built-in Explore, Plan agents
• ✅ Resume support - Can continue previous agent work
• ✅ Context passing - Findings flow between agents

---

Multi-Agent Architecture Theory

Why Multi-Agent Architectures

The Context Bottleneck: Single agents face inherent ceilings in reasoning capability, context management, and tool coordination. As tasks grow complex, context windows fill and performance degrades via the lost-in-middle effect, attention scarcity, and context poisoning. Multi-agent architectures partition work across multiple clean context windows.

Token Economics Reality:

| Architecture | Token Multiplier | Use Case | |--------------|------------------|----------| | Single agent chat | 1x baseline | Simple queries | | Single agent with tools | ~4x baseline | Tool-using tasks | | Multi-agent system | ~15x baseline | Complex research/coordination |

The Parallelization Argument: Many tasks contain parallelizable subtasks. Multi-agent architectures assign each to a dedicated agent with fresh context, reducing total time to the longest subtask rather than the sum.

The Specialization Argument: Different tasks benefit from different system prompts, tool sets, and context structures. Specialized agents carry only what they need.

Architectural Patterns (Detailed)

Supervisor/Orchestrator: Central agent delegates to specialists and synthesizes results. Best for tasks with clear decomposition and where human oversight is important. Risk: supervisor context becomes bottleneck; "telephone game" problem where supervisors paraphrase sub-agent responses incorrectly.

Telephone Game Fix: Implement a forward_message tool allowing sub-agents to pass responses directly to users without supervisor synthesis when appropriate.

Peer-to-Peer/Swarm: No central control; agents communicate directly via handoff mechanisms. Best for flexible exploration where rigid planning is counterproductive. Risk: coordination complexity and divergence without central state.

Hierarchical: Strategy layer (goals) -> Planning layer (decomposition) -> Execution layer (atomic tasks). Best for large-scale projects with clear hierarchical structure.

Context Isolation as Design Principle

The primary purpose of multi-agent architectures is context isolation. Three mechanisms:

• Full context delegation: Sub-agent gets complete context (max capability, defeats isolation purpose)
• Instruction passing: Sub-agent gets only instructions (maintains isolation, limits flexibility)
• File system memory: Agents read/write persistent storage (shared state without context bloat)

Consensus and Coordination

• Weighted Voting: Weight by confidence or domain expertise, not simple majority
• Debate Protocols: Adversarial critique yields higher accuracy than collaborative consensus
• Trigger-Based Intervention: Monitor for stalls and sycophancy (agents mimicking without unique reasoning)

Failure Modes and Mitigations

| Failure | Mitigation | |---------|-----------| | Supervisor Bottleneck | Output schema constraints; workers return distilled summaries; checkpointing | | Coordination Overhead | Clear handoff protocols; batch results; async communication | | Divergence | Clear objective boundaries; convergence checks; time-to-live limits | | Error Propagation | Validate outputs before passing; retry with circuit breakers; idempotent operations |

---

Multi-Agent Optimization

Performance Profiling

Profile across layers with specialized agents:

1. Database Performance Agent: Query execution time, index utilization, resource consumption
2. Application Performance Agent: CPU/memory profiling, algorithmic complexity, concurrency analysis
3. Frontend Performance Agent: Rendering metrics, network optimization, Core Web Vitals

Context Window Optimization

• Intelligent context compression with semantic relevance filtering
• Dynamic context window resizing and token budget management

Agent Coordination Efficiency

• Parallel execution design with minimal inter-agent communication overhead
• Dynamic workload distribution with fault-tolerant interactions

class MultiAgentOrchestrator:
    def __init__(self, agents):
        self.agents = agents
        self.execution_queue = PriorityQueue()
        self.performance_tracker = PerformanceTracker()

    def optimize(self, target_system):
        with concurrent.futures.ThreadPoolExecutor() as executor:
            futures = {
                executor.submit(agent.optimize, target_system): agent
                for agent in self.agents
            }
            for future in concurrent.futures.as_completed(futures):
                agent = futures[future]
                result = future.result()
                self.performance_tracker.log(agent, result)

Cost Optimization Strategies

• Token usage tracking and adaptive model selection
• Caching and result reuse across agents
• Dynamic model selection based on task complexity and budget

Latency Reduction

• Predictive caching and pre-warming agent contexts
• Intelligent result memoization
• Reduced round-trip communication

Monitoring and Continuous Improvement

• Real-time performance dashboards
• Automated optimization feedback loops
• Quality vs speed tradeoff management

Key Optimization Principles

• Always measure before and after optimization
• Maintain system stability during optimization
• Balance performance gains with resource consumption
• Implement gradual, reversible changes

Framework References

• LangGraph - Multi-agent patterns and state management
• AutoGen - GroupChat and conversational patterns
• CrewAI - Hierarchical agent processes