# What Reusable Artifacts Are Produced in Each Lesson's `outputs/` Directory?

> Discover reusable artifacts like Prompts, Skills, Agents, and MCP Servers generated in the outputs directory of AI Engineering from Scratch. Use them immediately in LLM chats, agent frameworks, or production services.

- Repository: [Rohit Ghumare/ai-engineering-from-scratch](https://github.com/rohitg00/ai-engineering-from-scratch)
- Tags: how-to-guide
- Published: 2026-06-21

---

**Each lesson's `outputs/` directory ships four concrete artifact types—Prompts, Skills, Agents, and MCP Servers—that you can immediately paste into LLM chats, load into agent frameworks, or deploy as production services.**

The `ai-engineering-from-scratch` curriculum standardizes every lesson around a predictable file structure where the `outputs/` folder serves as the delivery mechanism for production-ready components. According to the repository's README, these directories contain the tangible assets generated by each lesson, transforming theoretical concepts into immediately reusable code. Understanding what these artifacts are and how to consume them unlocks the curriculum's full value as a **503-component toolbox** for AI engineering.

## The Standardized Lesson Structure

Every lesson in the repository follows a uniform layout defined in the curriculum documentation:

```

phases/<NN>-<phase-name>/<NN>-<lesson-name>/
│
├── code/      # runnable implementations

├── docs/
│   └── en.md  # lesson narrative

└── outputs/   # reusable artifacts

```

The `outputs/` folder is the designated location where each lesson ships concrete, reusable artifacts. According to the source code analysis of the repository structure, this standardization ensures that across all phases—from deep learning fundamentals to agent engineering—you can predict exactly where to find the takeaway assets.

## The Four Types of Reusable Artifacts

Each `outputs/` directory contains one of four distinct artifact families, all using `.md` extensions for human readability and version control compatibility.

### Prompts (.md)

**Prompts** are ready-to-paste text files designed for use with any LLM assistant. These files contain carefully engineered prompt templates that you can copy directly into chat interfaces like Claude, ChatGPT, or Cursor.

Example file: [`phases/03-deep-learning-core/05-loss-functions/outputs/prompt-loss-function-selector.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/phases/03-deep-learning-core/05-loss-functions/outputs/prompt-loss-function-selector.md)

### Skills (.md)

**Skills** are tool-calling specifications that define capabilities for AI agents. These artifacts contain structured definitions that agents like Claude, Cursor, or OpenAI models can load to extend their functionality with specific tools or workflows.

Example file: [`phases/14-agent-engineering/10-skill-libraries-voyager/outputs/skill-voyager.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/phases/14-agent-engineering/10-skill-libraries-voyager/outputs/skill-voyager.md)

### Agents (.md)

**Agents** are minimal, self-contained agent implementations—often just a few dozen lines of Python—demonstrating specific capabilities like ReAct loops or tool use. Despite the `.md` extension, these files contain executable Python code that demonstrates the lesson's core concept in action.

Example file: [`phases/14-agent-engineering/01-the-agent-loop/outputs/agent-loop.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/phases/14-agent-engineering/01-the-agent-loop/outputs/agent-loop.md)

### MCP Servers (.md)

**MCP Servers** are Model-Context-Protocol servers that expose a model's context as a service. These artifacts allow you to integrate the lesson's functionality into production pipelines by running the server and connecting it to other tools via the MCP protocol.

Example file: [`phases/13-tools-and-protocols/07-building-an-mcp-server/outputs/mcp-server-example.md`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/phases/13-tools-and-protocols/07-building-an-mcp-server/outputs/mcp-server-example.md)

## How to Consume Artifacts from the outputs/ Directory

The repository provides consistent patterns for consuming each artifact type, whether through command-line utilities or Python APIs.

### Copying Prompts to Your Clipboard

For immediate use in LLM chat interfaces, copy prompt artifacts directly to your clipboard:

```bash
cat phases/03-deep-learning-core/05-loss-functions/outputs/prompt-loss-function-selector.md \
| pbcopy   # macOS – copy to clipboard

# Paste into your LLM UI

```

### Loading Skills into Agent Frameworks

Use the built-in skill loader to integrate skills into your agent runtime:

```python
from ai_engineering.skills import load_skill

skill = load_skill(
    "phases/14-agent-engineering/10-skill-libraries-voyager/outputs/skill-voyager.md"
)

result = skill.run(input_text="Summarize the lesson")
print(result)

```

### Running Agent Implementations

Execute agent artifacts directly using the repository's agent loader:

```python
from ai_engineering.agent import load_agent

agent = load_agent(
    "phases/14-agent-engineering/01-the-agent-loop/outputs/agent-loop.md"
)

response = agent.run("What is the ReAct loop?")
print(response)

```

### Launching MCP Servers

Start MCP server artifacts as standalone services with configurable parameters:

```bash
python phases/13-tools-and-protocols/07-building-an-mcp-server/outputs/mcp-server-example.md \
    --model gpt-4o-mini \
    --port 8080

```

## Automation and Installation Utilities

The repository includes automation scripts that eliminate manual artifact management:

- **[`scripts/install_skills.py`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/scripts/install_skills.py)** – Scans every `outputs/` directory across all lessons and installs skill artifacts into your environment, eliminating the need to manually copy individual files.

- **[`site/build.js`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/site/build.js)** – Generates the documentation website where each lesson's `outputs/` artifacts are displayed as downloadable links, making the 503-component collection browsable without cloning the repository.

## Summary

- **Four artifact types** exist in every lesson's `outputs/` directory: Prompts, Skills, Agents, and MCP Servers.
- **Consistent location** follows the pattern `phases/<NN>-<phase-name>/<NN>-<lesson-name>/outputs/` across the entire curriculum.
- **Immediate usability** requires no modification—artifacts are production-ready and runnable.
- **503 total components** comprise the complete toolbox available across all lessons.
- **Automated installation** via [`scripts/install_skills.py`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/scripts/install_skills.py) streamlines bulk deployment of skill artifacts.

## Frequently Asked Questions

### What file extension do the reusable artifacts use?

All artifacts use `.md` extensions, even when they contain Python code or configuration data. This design choice makes artifacts human-readable, diff-friendly in version control, and easy to preview in GitHub's interface while remaining executable or copy-paste ready.

### How do I install all skills from every lesson at once?

Run the [`scripts/install_skills.py`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/scripts/install_skills.py) utility, which recursively scans every `outputs/` directory in the repository and installs all skill artifacts into your environment. This eliminates manual copy-paste operations and ensures you have the complete skill library available for agent development.

### Can I use these artifacts with Claude or other AI assistants?

Yes. **Prompts** paste directly into any LLM chat interface, including Claude, ChatGPT, and Cursor. **Skills** load into Claude, Cursor, or OpenAI-compatible agents using the built-in `load_skill()` function from the `ai_engineering.skills` module. **Agents** run locally via Python, and **MCP Servers** integrate with any MCP-compatible client.

### Where can I view these artifacts without cloning the repository?

The [`site/build.js`](https://github.com/rohitg00/ai-engineering-from-scratch/blob/main/site/build.js) script generates a static documentation website where each lesson's `outputs/` artifacts appear as downloadable links. This allows you to browse the complete 503-component collection, preview artifact contents, and download specific files before deciding to clone the full repository.