# Build Reliable AI Agent Systems

AI agent projects often lack a structured path from architecture through evaluation. This workflow organizes design, implementation, orchestration, tools, memory, and testing into seven phases.

## Install

```bash
npx skillstore add sickn33/ai-agent-development
```

## Metadata

- Status: approved
- Slug: sickn33-ai-agent-development
- Skillstore revision: r2
- Version status: missing
- Tree hash: 2a195b506030b50629f04a5399a9f1be0da914f378adf271563930f67d21d369
- Author: sickn33
- GitHub username: sickn33
- License: MIT
- Repository: https://github.com/sickn33/antigravity-awesome-skills/tree/main/skills/ai-agent-development
- Ref: ebdfe608f5de2b66ff37ab4af12af8ac4f5e8006
- 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-ai-agent-development
- Manifest: https://skillstore.pages.dev/api/skills/sickn33-ai-agent-development/manifest

## Capabilities

- Structures AI agent work into seven phases from architecture through evaluation.
- Identifies supporting skills for autonomous agents, CrewAI, LangGraph, tools, memory, and testing.
- Provides reusable prompts for each major workflow phase.
- Defines steps for agent roles, communication, delegation, state management, and persistence.
- Covers tool interface design, error handling, memory types, and retrieval testing.
- Supplies an architecture map and a concise quality-gate checklist.

## Use Cases

- Plan a First Autonomous Agent: Turn a product goal into defined capabilities, tools, memory, safety boundaries, and success metrics.
- Coordinate a Multi-Agent Team: Define specialized roles, communication paths, delegation rules, and coordination tests for a CrewAI system.
- Build Stateful Agent Workflows: Plan LangGraph states, conditional branches, persistence, tool interfaces, memory, and evaluation scenarios.

## Prompt Templates

### Design a Single Agent

```
Design a single AI agent for [goal]. Define its users, inputs, outputs, capabilities, tools, memory, safety boundaries, and success metrics.
```

### Assign Multi-Agent Roles

```
Design a CrewAI system for [workflow]. Specify each role, responsibilities, shared context, delegation rules, communication paths, failure handling, and coordination tests.
```

### Orchestrate with LangGraph

```
Create a LangGraph workflow for [process]. Define state, nodes, transitions, conditional branches, checkpoints, retries, human approvals, persistence, and termination criteria.
```

### Evaluate and Harden an Agent

```
Evaluate [agent system] against [goals]. Create representative scenarios, edge cases, tool failures, safety tests, measurable criteria, and an improvement plan.
```

## Limitations

- Provides workflow guidance but does not include implementation code or framework configuration.
- Depends on referenced skills being available in the user environment.
- Does not validate generated agents against deployed systems, credentials, or production data.
- Requires adaptation for domain rules, permissions, safety boundaries, and infrastructure.

## Best Practices

- Define permissions, success metrics, and stop conditions before selecting tools or frameworks.
- Test tool failures, ambiguous inputs, unsafe requests, and human escalation paths.
- Evaluate each agent independently before testing coordination and end-to-end behavior.

## Anti Patterns

- Do not grant broad tool access without explicit permissions and validation.
- Do not add multiple agents when one agent and a deterministic workflow are sufficient.
- Do not rely on conversational memory for durable state or auditable records.

## Security Audit

- Audited at: 2026-07-23T20:35:41.636\+00:00
- Summary: All 32 static findings are false positives caused by Markdown backticks around skill names and fenced text examples in SKILL.md. The file contains workflow guidance only, with no executable Ruby, shell commands, prompt injection, or other semantic security concern.

## Stats

- Views: 108
- Downloads: 9
- Favorites: 0
- Popularity score: 0