# Design Reliable Autonomous AI Agents

Autonomous agent projects often fail through weak controls, unclear tools, and unsuitable architecture. This skill guides practical designs for planning, memory, recovery, orchestration, and observability.

## Install

```bash
npx skillstore add sickn33/ai-agents-architect
```

## Metadata

- Status: approved
- Slug: sickn33-ai-agents-architect
- Skillstore revision: r2
- Version status: missing
- Tree hash: c227d99346d34d2cfb3e7a482d9105e860ee8ca36bd0542ce0c90d0478a7c813
- Author: sickn33
- GitHub username: sickn33
- License: MIT
- Repository: https://github.com/sickn33/antigravity-awesome-skills/tree/main/skills/ai-agents-architect
- 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: 77
- Quality tier: bronze
- Public page: https://skillstore.pages.dev/skills/sickn33-ai-agents-architect
- Manifest: https://skillstore.pages.dev/api/skills/sickn33-ai-agents-architect/manifest

## Capabilities

- Selects between ReAct, plan-and-execute, supervisor, and tool registry patterns.
- Defines iteration limits, timeouts, cost caps, and tool failure controls.
- Structures working, episodic, and semantic memory with retrieval guidance.
- Designs focused tool specifications with parameters, examples, and expected errors.
- Plans checkpoint recovery, tracing, evaluation, and debugging for long-running agents.
- Assesses when specialist agents justify multi-agent coordination overhead.

## Use Cases

- Design a First Tool-Using Agent: Choose an agent loop, define focused tools, and add practical limits for a bounded application workflow.
- Plan a Long-Running Agent Platform: Structure memory, checkpoints, recovery, and observability for agents that maintain task progress across failures.
- Review Agent Reliability: Identify uncontrolled loops, silent tool errors, excessive memory, brittle parsing, and unnecessary multi-agent complexity.

## Prompt Templates

### Choose an Agent Pattern

```
Design an AI agent for [task]. Compare ReAct and plan-and-execute, then recommend one with required tools, limits, and failure conditions.
```

### Specify Tools and Controls

```
Create a tool plan for an agent that handles [workflow]. Define each tool purpose, parameters, errors, retries, permissions, and execution limits.
```

### Design Memory and Recovery

```
Architect memory and checkpoint recovery for [long-running task]. Separate working, episodic, and semantic memory, with retention, retrieval, cleanup, and resumption rules.
```

### Evaluate a Multi-Agent System

```
Review this proposed multi-agent system: [architecture]. Assess role boundaries, handoffs, supervisor logic, failure recovery, observability, cost, and whether one agent is sufficient.
```

## Limitations

- Provides architecture guidance, not a complete runnable agent implementation.
- Does not replace framework documentation, environment testing, or security review.
- Requires clear goals, available tools, permissions, and operational constraints.
- Does not provide benchmark results or guarantee agent reliability.

## Best Practices

- Set iteration, token, time, and cost limits before enabling autonomous execution.
- Give every tool a clear purpose, typed parameters, examples, permissions, and recoverable error responses.
- Trace decisions, tool calls, latency, token usage, checkpoints, and final outcomes.

## Anti Patterns

- Do not run agent loops without explicit stopping conditions and circuit breakers.
- Do not store every observation or expose large tool catalogs without relevance filtering.
- Do not introduce multiple agents before one well-equipped agent demonstrates measurable limitations.

## Security Audit

- Audited at: 2026-07-23T20:38:42.746\+00:00
- Summary: Both static findings are false positives caused by ordinary prose and inline Markdown identifiers. The skill contains architectural guidance without executable commands, network reconnaissance, or semantic threats.

## Stats

- Views: 32
- Downloads: 12
- Favorites: 0
- Popularity score: 0