# Design Reliable AI Context Systems

AI applications lose relevance when context, memory, and retrieval systems are poorly designed. This skill guides context architecture, RAG, knowledge graphs, and multi-agent state management.

## Install

```bash
npx skillstore add sickn33/context-manager
```

## Metadata

- Status: approved
- Slug: sickn33-context-manager
- Skillstore revision: r2
- Version status: missing
- Tree hash: 208acfc4d6e2719c1e590b8fc251ce75c768f9b286e91c1f3add7c6156875487
- Author: sickn33
- GitHub username: sickn33
- License: MIT
- Repository: https://github.com/sickn33/antigravity-awesome-skills/tree/main/skills/context-manager
- Ref: f9e2c34b4f19c7f3e6b0a1e93227b5f77cc12526
- 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/sickn33-context-manager
- Manifest: https://skillstore.pages.dev/api/skills/sickn33-context-manager/manifest

## Capabilities

- Designs dynamic context assembly, pruning, relevance filtering, and token budget strategies.
- Recommends vector database, embedding, semantic search, and hybrid retrieval approaches.
- Structures RAG pipelines with chunking, query understanding, ranking, and freshness controls.
- Plans persistent, episodic, semantic, and working memory architectures.
- Defines multi-agent context routing, handoffs, state management, and conflict resolution.
- Proposes quality, latency, cost, relevance, and staleness metrics for context systems.

## Use Cases

- Plan a Support Context Architecture: Design retrieval, memory, and agent handoffs for consistent responses across long-running customer cases.
- Improve Enterprise Document Retrieval: Select chunking, embedding, hybrid search, ranking, and freshness strategies for a large internal document collection.
- Control Multi-Agent Workflow State: Define context ownership, routing, handoff, and recovery rules for a coordinated AI workflow.

## Prompt Templates

### Assess Basic Context Needs

```
Review this AI workflow: [workflow]. Identify required context, missing inputs, token constraints, and a simple assembly strategy.
```

### Design a Retrieval Pipeline

```
Design a RAG pipeline for [corpus]. Recommend chunking, embeddings, retrieval, ranking, freshness checks, and evaluation metrics.
```

### Plan Long-Term Memory

```
Create a memory strategy for [application]. Separate working, episodic, semantic, and persistent memory with retention and retrieval rules.
```

### Architect Multi-Agent Context

```
Architect context management for [agents and workflow]. Define ownership, handoffs, isolation, token budgets, failure recovery, governance, and measurable service targets.
```

## Limitations

- Provides architecture guidance but does not deploy databases, models, APIs, or monitoring systems.
- Requires environment-specific requirements, permissions, testing, and expert review before implementation.
- Does not independently access enterprise data sources or validate live retrieval performance.
- The referenced implementation playbook is not included in the reviewed package.

## Best Practices

- Define measurable relevance, latency, cost, freshness, and privacy requirements before selecting components.
- Separate durable facts, conversation history, retrieved evidence, and temporary working context.
- Evaluate retrieval and final responses with representative tasks, failure cases, and stale data.

## Anti Patterns

- Do not place every available document into each prompt.
- Do not choose an embedding model or vector database without measured requirements.
- Do not store sensitive conversation history without retention, access, and deletion controls.

## Security Audit

- Audited at: 2026-07-23T21:17:01.162\+00:00
- Summary: Both static findings are false positives. The cited lines contain a Markdown file reference and a hybrid-search capability statement, without shell execution or system reconnaissance. No semantic security issue was found.

## Stats

- Views: 143
- Downloads: 19
- Favorites: 0
- Popularity score: 0