# Improve AI Prompts Across Text, Image, and Video

Many prompts produce vague or inconsistent model results. This skill teaches structured prompts, examples, constraints, and model-specific wording for Claude, Codex, images, and video.

## Install

```bash
npx skillstore add inference-sh/prompt-engineering
```

## Metadata

- Status: approved
- Slug: inference-sh-prompt-engineering
- Skillstore revision: r1
- Version status: missing
- Tree hash: 6d2edcdafb874a89769547754fcb33a9ae17ff72df2a2c7f8f7dc8159f230d8c
- Author: inference-sh
- GitHub username: inference-sh
- License: MIT
- Repository: https://github.com/inference-sh/skills/tree/main/skills/prompt-engineering/
- Ref: a06681402992ceae98ba04d54cfd4ab004862696
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: blocked
- Manual install advisory: allowed\_with\_warning
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, network
- Quality score: 38
- Quality tier: warning
- Public page: https://skillstore.pages.dev/skills/inference-sh-prompt-engineering
- Manifest: https://skillstore.pages.dev/api/skills/inference-sh-prompt-engineering/manifest

## Capabilities

- Explains a reusable prompt structure with role, task, constraints, and output format.
- Shows role prompting, task clarity, chain-of-thought, few-shot examples, and structured output requests.
- Provides image prompt guidance for subject detail, style, composition, quality terms, and negative prompts.
- Provides video prompt guidance for shot type, subject action, setting, style, and camera movement.
- Compares prompt tips for Claude, GPT-4, FLUX, and Veo style workflows.
- Includes reusable prompt templates for code review, content writing, and image generation.

## Use Cases

- Improve Team Prompt Quality: Create shared prompt patterns for support, analysis, writing, and coding tasks.
- Design Better Image Prompts: Write prompts with clearer subjects, style cues, composition, and negative prompt terms.
- Build Reliable AI Workflows: Use examples, constraints, and output formats to make repeated model runs more predictable.

## Prompt Templates

### Clarify a Simple Task

```
Rewrite my prompt for clarity. Goal: [goal]. Audience: [audience]. Constraints: [constraints]. Desired output: [format].
```

### Create a Few-Shot Prompt

```
Create a prompt with three examples. Task: [task]. Good outputs should show [qualities]. Bad outputs should avoid [issues].
```

### Design a Generation Prompt

```
Write a generation prompt for [medium]. Include subject, setting, style, composition, lighting, motion, and negative terms where useful.
```

### Refine a Production Prompt

```
Audit this prompt for ambiguity, missing context, conflicting constraints, safety concerns, and output format risks. Then provide a revised version.
```

## Limitations

- The skill is a guide and does not guarantee model accuracy or safety.
- Examples depend on inference.sh models and may need account setup.
- It does not validate prompts against policy, privacy, or copyright requirements.
- Some examples include command-line usage that should be reviewed before execution.

## Best Practices

- State the role, task, constraints, and output format before adding examples.
- Use concrete examples when style, tone, or classification rules matter.
- Review command examples before running them in a local or shared environment.

## Anti Patterns

- Asking broad questions without context, audience, or success criteria.
- Combining conflicting instructions that make the desired output unclear.
- Running installer commands from a guide without checking the source and integrity.

## Security Audit

- Audited at: 2026-07-06T16:04:09.673\+00:00
- Summary: Most static shell-execution findings are false positives caused by Markdown code fences and prompt examples, not executable Ruby backticks. The line 21 install instruction is confirmed critical because it pipes a remote script directly into sh; the related hardcoded installer URL is also confirmed. No additional semantic prompt-injection or data-exfiltration intent was found.

## Stats

- Views: 218
- Downloads: 8
- Favorites: 0
- Popularity score: 0