# Analyze Agent Memory Architecture

Agent frameworks often hide how context, memory, and token budgets interact. This skill gives Claude, Codex, and Claude Code a structured way to inspect memory flow and compare design choices.

## Install

```bash
npx skillstore add dowwie/memory-orchestration
```

## Metadata

- Status: approved
- Slug: dowwie-memory-orchestration
- Skillstore revision: r1
- Version status: missing
- Tree hash: 39f4393a3e0286d4970f1ea473feafb5f03a081f647996032252fb0b019babd5
- Author: Dowwie
- GitHub username: Dowwie
- License: MIT
- Repository: https://github.com/Dowwie/agent\_framework\_study/tree/main/.claude/skills/memory-orchestration
- 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-memory-orchestration
- Manifest: https://skillstore.pages.dev/api/skills/dowwie-memory-orchestration/manifest

## Capabilities

- Breaks down standard prompt assembly order for agent systems.
- Compares template, chat message, and programmatic assembly patterns.
- Evaluates FIFO, sliding window, summarization, vector, and importance-based eviction policies.
- Maps working, session, and persistent memory tiers.
- Reviews token counting methods and budget allocation choices.
- Provides a structured report format for memory architecture analysis.

## Use Cases

- Review Agent Framework Memory: Trace how an agent framework builds prompts, stores history, and handles context overflow.
- Compare Context Strategies: Compare memory tiers, eviction policies, and token allocation across multiple agent designs.
- Document State Continuity Risks: Create a clear report on where user preferences, summaries, or retrieved memories may be lost.

## Prompt Templates

### Trace Prompt Assembly

```
Use memory-orchestration to inspect this agent framework. Identify how the system prompt, tools, history, memory, and current input are assembled.
```

### Evaluate Context Overflow

```
Analyze the context overflow strategy. Identify the eviction trigger, what content is removed, and what user experience risks it creates.
```

### Map Memory Tiers

```
Map the working, session, and persistent memory tiers. Explain storage location, retrieval method, capacity assumptions, and failure risks.
```

### Compare Production Strategies

```
Compare the memory orchestration strategies in these agent systems. Rank prompt assembly, eviction, persistence, and token budget choices by operational risk.
```

## Limitations

- Provides analysis guidance, not runnable tooling or automatic code scanning.
- Examples are illustrative and must be adapted to each framework.
- Does not benchmark real retrieval quality or model performance.
- Does not implement databases, vector stores, or tokenizers.

## Best Practices

- Review actual prompt assembly paths before judging memory behavior.
- Separate short-term conversation history from persistent user memory.
- Reserve output tokens before filling the context window with history.

## Anti Patterns

- Assuming recent messages are always the most important context.
- Storing all conversation history without retention or privacy rules.
- Using summaries without checking what details they remove.

## Security Audit

- Audited at: 2026-07-06T12:17:18.974\+00:00
- Summary: All static external command findings are false positives from Markdown fences and inline backticks in SKILL.md. No evidence found for executable command use, network access, credential handling, or prompt injection attempts.

## Stats

- Views: 234
- Downloads: 10
- Favorites: 0
- Popularity score: 0