# Build Production-Ready AI Products

AI prototypes often fail under production traffic, unreliable outputs, prompt attacks, and rising costs. This skill provides practical patterns for reliable, secure, and measurable AI features.

## Install

```bash
npx skillstore add sickn33/ai-product
```

## Metadata

- Status: approved
- Slug: sickn33-ai-product
- Skillstore revision: r2
- Version status: missing
- Tree hash: 3a8f95a6003ef69a11ec635d8327705a484ee8106fb6c7d74f6d1eceda1330eb
- Author: sickn33
- GitHub username: sickn33
- License: MIT
- Repository: https://github.com/sickn33/antigravity-awesome-skills/tree/main/skills/ai-product
- 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, network
- Quality score: 76
- Public page: https://skillstore.pages.dev/skills/sickn33-ai-product
- Manifest: https://skillstore.pages.dev/api/skills/sickn33-ai-product/manifest

## Capabilities

- Designs LLM integrations with schema validation, error handling, and graceful fallback behavior.
- Explains hybrid RAG retrieval using semantic search, keyword search, reranking, and context limits.
- Provides prompt versioning, regression testing, and input separation practices.
- Plans streaming responses, asynchronous jobs, and resilient provider fallbacks.
- Identifies token, rate-limit, caching, and usage-tracking controls for cost management.

## Use Cases

- Plan a New AI Feature: Create an architecture covering model calls, validation, latency, failure handling, and cost controls before implementation.
- Harden an AI Prototype: Review a working prototype for prompt injection, malformed outputs, missing limits, weak monitoring, and provider failures.
- Design a RAG System: Define retrieval, reranking, context budgeting, source verification, and evaluation for a knowledge assistant.

## Prompt Templates

### Outline an AI Feature

```
Outline an AI feature for [user need]. Include inputs, outputs, model role, validation, latency target, fallback, and success metrics.
```

### Review Production Readiness

```
Review this AI workflow: [workflow]. Identify reliability, security, latency, cost, and observability gaps. Prioritize practical fixes.
```

### Design a RAG Architecture

```
Design RAG for [knowledge source] and [query type]. Specify ingestion, retrieval, reranking, context limits, citations, evaluation, and update strategy.
```

### Create an AI Operations Plan

```
Create an operations plan for [AI service]. Cover rate limits, budgets, caching, timeouts, fallbacks, monitoring, incident handling, and prompt versioning.
```

## Limitations

- Provides architectural guidance and examples, but does not deploy or operate infrastructure.
- Does not select a model without workload, quality, latency, and budget requirements.
- Example code requires adaptation, dependency review, testing, and security validation.
- Does not replace legal, medical, financial, privacy, or domain expert review.

## Best Practices

- Validate model outputs and treat all generated content as untrusted input.
- Version prompts, test representative cases, and measure quality after every change.
- Track latency, token usage, cost, failures, and user-level limits from the first release.

## Anti Patterns

- Do not place unsanitized user input inside privileged prompt instructions.
- Do not use the largest model for every task without measured quality benefits.
- Do not deploy an AI dependency without timeouts, fallbacks, monitoring, and graceful error states.

## Security Audit

- Audited at: 2026-07-23T20:46:19.508\+00:00
- Summary: All 35 static findings are false positives caused by TypeScript template literals, Markdown fences, ordinary prose, or quoted defensive examples. The skill contains educational guidance and illustrative code, with no executable shell commands, credential access, reconnaissance, or malicious prompt injection.

## Stats

- Views: 97
- Downloads: 10
- Favorites: 1
- Popularity score: 0