# Design Agent Framework Architectures

Designing an agent framework requires clear primitives, interfaces, and execution rules. This skill turns prior analysis into a reference architecture and implementation roadmap.

## Install

```bash
npx skillstore add dowwie/architecture-synthesis
```

## Metadata

- Status: approved
- Slug: dowwie-architecture-synthesis
- Skillstore revision: r1
- Version status: missing
- Tree hash: a3c5d9bac27b61e7e46ecf7d90d84b650c9eee549581462194b1a4c25d091f9d
- Author: Dowwie
- GitHub username: Dowwie
- License: MIT
- Repository: https://github.com/Dowwie/agent\_framework\_study/tree/main/.claude/skills/architecture-synthesis
- Ref: 72d5025b022c77f7a51bdf5c1637c689c80e89d1
- 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
- Quality score: 78
- Quality tier: bronze
- Public page: https://skillstore.pages.dev/skills/dowwie-architecture-synthesis
- Manifest: https://skillstore.pages.dev/api/skills/dowwie-architecture-synthesis/manifest

## Capabilities

- Defines Message, AgentState, ToolResult, AgentFinish, and AgentContinue primitives.
- Specifies LLM, Tool, and Memory protocols with clear method responsibilities.
- Outlines an agent execution loop with parsing, tool execution, events, and step limits.
- Provides a Mermaid architecture diagram for core, protocol, execution, and integration layers.
- Creates phased implementation roadmap and expected documentation artifacts.

## Use Cases

- Plan a New Agent Framework: Turn framework research into a concrete architecture with primitives, interfaces, and execution flow.
- Standardize Agent Interfaces: Define shared contracts for LLM calls, tools, memory, results, and callbacks before implementation.
- Prepare Implementation Work: Convert design decisions into phased milestones, documentation artifacts, and example deliverables.

## Prompt Templates

### Create a Basic Architecture

```
Use architecture-synthesis to draft a reference architecture for a simple agent framework. Include primitives, protocols, loop design, and roadmap.
```

### Compare Requirements to Architecture

```
Use architecture-synthesis with my requirements and comparative matrix. Highlight which primitives, interfaces, and execution rules satisfy each requirement.
```

### Design Extension Points

```
Use architecture-synthesis to define extension points for tools, memory, callbacks, streaming, and error handling in an agent framework.
```

### Synthesize Production Roadmap

```
Use architecture-synthesis to create a production-ready architecture brief. Include provider integrations, observability, failure modes, implementation phases, and validation criteria.
```

## Limitations

- Does not inspect a codebase or generate runnable project files by itself.
- Assumes comparative analysis, anti-patterns, and requirements are already available.
- Examples are Python-oriented and may need adaptation for other languages.
- Does not validate provider APIs, credentials, or production infrastructure.

## Best Practices

- Provide the comparative matrix, anti-pattern catalog, and design requirements before using the skill.
- Review generated primitives and protocols with implementers before starting production code.
- Treat diagrams and roadmap phases as planning artifacts, then validate them against real constraints.

## Anti Patterns

- Do not use the architecture as a security review or production readiness assessment.
- Do not skip requirements and ask for a full framework design from vague goals.
- Do not assume the Python examples directly fit every runtime or language.

## Security Audit

- Audited at: 2026-07-06T11:56:35.546\+00:00
- Summary: All static findings are false positives caused by Markdown code fences, inline backticks, and Mermaid artifact names. SKILL.md contains architecture templates and diagrams only; I found no executable shell commands, reconnaissance logic, exfiltration intent, or prompt injection text.

## Stats

- Views: 225
- Downloads: 6
- Favorites: 0
- Popularity score: 0