# Analyze Agent Framework Architecture

Agent framework design is hard to compare because behavior spans code structure and reasoning patterns. This skill guides Claude, Codex, and Claude Code through a staged architecture forensics workflow.

## Install

```bash
npx skillstore add dowwie/architectural-forensics
```

## Metadata

- Status: approved
- Slug: dowwie-architectural-forensics
- Skillstore revision: r1
- Version status: missing
- Tree hash: 7e99d8b7867e4869fa43833665c0cddefb1f4d1a92ec31f81d43a6eb90d37b67
- Author: Dowwie
- GitHub username: Dowwie
- License: MIT
- Repository: https://github.com/Dowwie/agent\_framework\_study/tree/main/.claude/skills/architectural-forensics
- 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: filesystem, external\_commands
- Quality score: 76
- Public page: https://skillstore.pages.dev/skills/dowwie-architectural-forensics
- Manifest: https://skillstore.pages.dev/api/skills/dowwie-architectural-forensics/manifest

## Capabilities

- Coordinates phased analysis of engineering chassis, cognitive architecture, and synthesis outputs.
- Uses codebase maps and clustered file reading to keep agent context bounded.
- Defines roles for orchestrator, framework, skill, reader, and synthesis agents.
- Produces framework summaries, comparison matrices, anti-pattern catalogs, and reference architecture drafts.
- Supports resumable work with manifest and per-framework state file guidance.

## Use Cases

- Evaluate Framework Candidates: Compare agent frameworks before adopting one for a platform or product team.
- Design A New Agent Runtime: Extract reusable patterns and avoid known architecture problems during system design.
- Document Technical Due Diligence: Create evidence-based architecture reports for acquisition, vendor, or open source reviews.

## Prompt Templates

### Map One Framework

```
Analyze the agent framework at [path]. Create a concise architecture summary with engineering chassis, cognitive architecture, and key tradeoffs.
```

### Compare Two Frameworks

```
Compare [framework A] and [framework B]. Focus on execution engines, memory, tool interfaces, and error handling. Recommend patterns to reuse.
```

### Audit One Subsystem

```
Study the tool interface in [framework]. Cite files and summarize schema generation, dispatch behavior, error feedback, and extension points.
```

### Build Reference Architecture

```
Analyze [framework list], build a comparison matrix, identify anti-patterns, and draft a reference architecture for a new agent framework.
```

## Limitations

- Requires access to the target repository and related analysis skills.
- Does not include the external codebase mapping script referenced in the workflow.
- Findings depend on source quality, repository size, and generated map accuracy.
- Manual review is still needed for framework-specific design tradeoffs.

## Best Practices

- Start with a fresh codebase map so agents analyze current files.
- Keep framework outputs separated to preserve evidence and avoid cross-contamination.
- Review generated recommendations before turning them into implementation plans.

## Anti Patterns

- Do not use the synthesis phase before individual framework summaries are complete.
- Do not treat generated architecture advice as a substitute for source review.
- Do not compare frameworks without normalizing versions, languages, and use cases.

## Security Audit

- Audited at: 2026-07-06T11:52:44.462\+00:00
- Summary: No confirmed security issues were found in the reviewed files. The static alerts are Markdown examples, bounded workflow output paths, benign field names, or report headings. No prompt injection or data-exfiltration intent was found.

## Stats

- Views: 475
- Downloads: 8
- Favorites: 2
- Popularity score: 0