deploy-multi-agent-swarm
/deploy-multi-agent-swarm
Wdraża rój wielu agentów: zdecentralizowaną topologię, protokół licytacji, głosowanie konsensusu, wykonanie równoległe, rozwiązywanie konfliktów i magazyn stanu roju.
Ocena i bezpieczeństwo
- Pliki w paczce
- 3
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 26.10.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
Co robi i kiedy
Wdraża rój wielu agentów: zdecentralizowaną topologię, protokół licytacji, głosowanie konsensusu, wykonanie równoległe, rozwiązywanie konfliktów i magazyn stanu roju.
Oryginalny opis (EN)
Deploy Multi-Agent Swarm — configure decentralized agent topology, bidding protocol, consensus voting, parallel execution, conflict resolution, swarm state store. Use when: deploy, provision swarm.
Jak działa
- When to Use
- How Swarm Differs from Multi-Agent Service (Play 07)
- Prerequisites
- Step 1: Deploy Infrastructure
- Step 2: Configure Swarm Topology
- Step 3: Configure Bidding Protocol
- Step 4: Configure Consensus Mechanism
- Step 5: Deploy Swarm State Store
- Step 6: Post-Deployment Verification
- Troubleshooting
Podgląd SKILL.md
---
name: deploy-multi-agent-swarm
description: "Deploy Multi-Agent Swarm — configure decentralized agent topology, bidding protocol, consensus voting, parallel execution, conflict resolution, swarm state store. Use when: deploy, provision swarm."
---
# Deploy Multi-Agent Swarm
## When to Use
- Deploy a decentralized multi-agent system (no single supervisor)
- Configure bidding protocol for task assignment
- Set up consensus voting for quality decisions
- Enable parallel execution with conflict resolution
- Deploy swarm state coordination (Redis/Cosmos DB)
## How Swarm Differs from Multi-Agent Service (Play 07)
| Aspect | Play 07 (Supervisor) | Play 22 (Swarm) |
|--------|---------------------|------------------|
| Coordination | Central supervisor decides | Agents bid and vote |
| Task assignment | Top-down delegation | Competitive bidding |
| Quality control | Supervisor aggregates | Consensus voting |
| Failure handling | Supervisor reassigns | Swarm self-heals |
| Scalability | Limited by supervisor | Horizontally scalable |
| Single point of failure | Supervisor is SPOF | No SPOF |
## Prerequisites
1. Azure CLI authenticated: `az account show`
2. Azure OpenAI with multiple deployments (agents need separate context)
3. Azure Service Bus (agent-to-agent pub/sub messaging)
4. Redis or Cosmos DB (swarm state coordination)
5. Container Apps (per-agent container hosting)
## Step 1: Deploy Infrastructure
```bash
az bicep lint -f infra/main.bicep
az deployment group create -g $RG -f infra/main.bicep -p infra/parameters.json
```
## Step 2: Configure Swarm Topology
```json
{
"swarm": {
"agents": [
{ "role": "researcher", "model": "gpt-4o-mini", "instances": 3 },
{ "role": "analyst", "model": "gpt-4o", "instances": 2 },
{ "role": "writer", "model": "gpt-4o-mini", "instances": 2 },
{ "role": "validator", "model": "gpt-4o-mini", "instances": 2 }
],
"bidding": { "timeout_ms": 5000, "min_bids": 2 },
"consensus": { "method": "majority_vote", "min_votes": 3, "quorum": 0.6 },
"max_parallel": 5,
"max_rounds": 3
}
}
```
## Step 3: Configure Bidding Protocol
| Phase | Duration | What Happens |
|-------|----------|-------------|
| Task broadcast | 0s | Task published to all agents via Service Bus |
| Bidding window | 0-5s | Agents assess task, submit bid with confidence score |
| Selection | 5s | Top N bids selected (by confidence × past accuracy) |
| Execution | 5-30s | Selected agents execute in parallel |
| Voting | 30-35s | All agents vote on best result |
| Consensus | 35s | Majority-approved result returned |
## Step 4: Configure Consensus Mechanism
| Method | Min Agents | Use Case |
|--------|-----------|----------|
| Majority vote | 3+ | General-purpose decisions |
| Weighted vote | 3+ | Expert agents get more weight |
| Unanimous | All | Safety-critical decisions |
| First-past-threshold | 1+ | Speed-optimized (accept first good-enough) |
## Step 5: Deploy Swarm State Store
- Redis for fast in-flight state (bids, votes, partial results)
- Cosmos DB for durable history (task logs, agent performance)
- Key pattern: `swarm:{task_id}:bids`, `swarm:{task_id}:votes`
## Step 6: Post-Deployment Verification
- [ ] Agents register in swarm (visible in agent registry)
- [ ] Bidding protocol working (agents bid within timeout)
- [ ] Consensus voting producing majority-accepted results
- [ ] Parallel execution not exceeding max_parallel
- [ ] Conflict resolution handling disagreements
- [ ] Swarm self-healing when an agent fails
- [ ] Cost per swarm task tracked
## Troubleshooting
| Issue | Cause | Fix |
|-------|-------|-----|
| No bids received | Agents not subscribed to topic | Check Service Bus subscriptions |
| Always same agent wins | Bidding not diverse | Rotate winners, add cooldown |
| Consensus never reached | Quorum too high (1.0) | Lower to 0.6 |
| Conflicting results | No conflict resolution | Add validator vote as tiebreaker |
| High cost | Too many agents bidding | Reduce swarm size, use fewer roles |
| Slow response | Sequential voting | Parallelize bid + vote phases |
## Architecture Diagram
```
Task → Service Bus (broadcast) → All Agents (bid) → Selection → Parallel Execution
↓
Results → Consensus Vote → Best Output
↓
Disagreement? → Extra Round or Human Escalation
```
Pliki w paczce (3)
- SKILL.lean.md 4 KB
- SKILL.md 4 KB
- agents/openai.yaml 1 KB
Wersje
W bazie nie ma innych skilli o nazwie deploy-multi-agent-swarm.
Migawka bazy z 13.11.2026 · wersja paczki 45d03f7d9b10 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100evaluate-multi-agent-swarm
Ocenia rój wielu agentów: jakość konsensusu, efektywność licytacji, zgodność głosów, przyspieszenie równoległe i koszt zadania roju. - 100tune-multi-agent-swarm
Dostraja rój agentów: rozmiar roju, strategię licytacji, progi głosowania, dobór modeli i limity równoległego wykonania. - 90swarm-consensus
Wprowadza głosowanie, większościowe lub ważone, do rozstrzygania konfliktów między agentami. - 90autonomous-swarm-patterns
Projektuje samoorganizujące się roje autonomicznych agentów, w tym mechanizmy konsensusu. - 98swarm-architectures
Prototypuje architektury roju agentów AI: router do specjalistów, rundy debaty z sędzią i pomiar, czy rój przewyższa pojedynczy model. - 100multi-agent-patterns
Pomaga projektować systemy wieloagentowe: wzorzec supervisor, architektura swarm, koordynacja wielu agentów, izolacja kontekstu, przekazania, subagenci i równoległe wykonanie.