# Design Reliable Multi-Agent Systems

Complex agent workflows can lose context, diverge, or waste tokens through poor coordination. This skill provides practical architecture patterns, handoff protocols, and failure controls.

## Install

```bash
npx skillstore add sickn33/multi-agent-patterns
```

## Metadata

- Status: approved
- Slug: sickn33-multi-agent-patterns
- Skillstore revision: r2
- Version status: missing
- Tree hash: a0ad39e606ddcb903900ca8a57936677a29d271ae8dd79a2cc0a2c5cde33c76d
- Author: sickn33
- GitHub username: sickn33
- License: MIT
- Repository: https://github.com/sickn33/antigravity-awesome-skills/tree/main/skills/multi-agent-patterns
- Ref: 81e05e636292629114b76cbb3922fbe57672fc02
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network
- Quality score: 78
- Quality tier: bronze
- Public page: https://skillstore.pages.dev/skills/sickn33-multi-agent-patterns
- Manifest: https://skillstore.pages.dev/api/skills/sickn33-multi-agent-patterns/manifest

## Capabilities

- Compares supervisor, peer-to-peer, and hierarchical multi-agent architectures.
- Explains how separate agent contexts support parallel work and specialization.
- Defines direct forwarding and explicit handoff patterns for preserving response fidelity.
- Describes weighted voting, debate protocols, and trigger-based coordination.
- Provides mitigations for bottlenecks, divergence, communication overhead, and error propagation.

## Use Cases

- Plan a research agent team: Design parallel research, analysis, verification, and writing roles with clear aggregation rules.
- Choose an orchestration pattern: Compare centralized supervision, direct handoffs, and layered execution for a planned product workflow.
- Review coordination failures: Identify context bottlenecks, divergence risks, and missing validation controls in an existing agent design.

## Prompt Templates

### Select a basic pattern

```
Recommend a multi-agent pattern for [task]. Explain why it fits, define each agent, and describe the final result flow.
```

### Design agent handoffs

```
Design handoffs for [workflow]. Define routing conditions, required state, output validation, fallback behavior, and termination rules.
```

### Evaluate architecture tradeoffs

```
Compare supervisor, peer-to-peer, and hierarchical designs for [system]. Assess latency, token cost, context isolation, oversight, scalability, and failure propagation.
```

### Audit a production design

```
Review this multi-agent design: [design]. Find bottlenecks, unsafe handoffs, divergence paths, weak consensus, missing observability, and error propagation risks. Propose measurable controls.
```

## Limitations

- Provides architecture guidance rather than a complete framework implementation.
- Does not benchmark designs against a specific workload or production environment.
- Does not configure authentication, permissions, storage, or deployment infrastructure.
- Requires environment-specific testing before production use.

## Best Practices

- Give each agent a narrow objective, required inputs, output contract, and termination condition.
- Validate outputs before handoff and preserve full responses when supervisor summarization would remove important detail.
- Measure token use, latency, retries, handoff failures, and progress toward shared goals.

## Anti Patterns

- Do not create multiple agents when one agent can complete the task within its context.
- Do not let agents hand off work without explicit state, ownership, and convergence rules.
- Do not use simple majority voting when agents have unequal expertise or share the same weak evidence.

## Security Audit

- Audited at: 2026-08-04T15:10:36.081\+00:00
- Summary: All 19 static findings are false positives caused by Markdown fences, benign Python examples, ordinary prose, and documentation links. No command execution, unsolicited network request, reconnaissance, prompt injection, or data-exfiltration intent appears in SKILL.md.

## Stats

- Views: 125
- Downloads: 11
- Favorites: 0
- Popularity score: 0