Copilot Customization Reference Solution

This repository is a reference implementation for a UD01 activity on GitHub Copilot customizations (module: Programacion de Inteligencia Artificial, 5073).

Learners can use this project as a model answer: every customization primitive is represented with a realistic, justified example from an AI programming context.

Purpose

Demonstrate, in a working repository, how each of the five Copilot customization layers works, when to use them, and how they differ from the alternatives.

Project Structure

.
├── .github/
│   ├── copilot-instructions.md          # Layer 1: Global always-on instructions
│   ├── instructions/
│   │   ├── python.instructions.md       # Layer 1: Rules for .py files (applyTo)
│   │   ├── markdown.instructions.md     # Layer 1: Rules for .md files (applyTo)
│   │   └── notebooks.instructions.md   # Layer 1: Rules for .ipynb files (applyTo)
│   ├── prompts/
│   │   ├── tutor-review.prompt.md       # Layer 2: Code review via tutor agent
│   │   ├── sdd-check.prompt.md          # Layer 2: Spec-driven implementation check
│   │   ├── todo-to-plan.prompt.md       # Layer 2: Scan TODOs → prioritised plan
│   │   └── arch-review.prompt.md        # Layer 2: Architecture review
│   ├── skills/
│   │   ├── csv-eda-basica/              # Layer 3: EDA on a CSV + notebook template
│   │   ├── revision-notebook/           # Layer 3: Notebook quality audit
│   │   ├── dataset-card/                # Layer 3: Dataset documentation card
│   │   ├── debug-python-basico/         # Layer 3: Structured Python debugging
│   │   ├── todo-a-plan/                 # Layer 3: TODOs → backlog
│   │   ├── spec-a-tareas/               # Layer 3: Spec → task list (SDD)
│   │   ├── preparar-practica/           # Layer 3: Design a practice exercise
│   │   ├── conventional-commit/         # Layer 3: Git commit messages
│   │   ├── comparar-primitivas/         # Layer 3: Meta — choose the right primitive
│   │   ├── evaluar-contexto-necesario/  # Layer 3: Meta — what context to attach
│   │   └── frontmatter-designer/        # Layer 3: Meta — design frontmatter
│   ├── agents/
│   │   └── tutor.agent.md               # Layer 4: Socratic teaching agent
│   └── hooks/
│       ├── notebook-guardian.json       # Layer 5: PreToolUse hook config
│       └── scripts/
│           └── notebook-guardian.py     # Layer 5: Hook implementation script
├── actividade.md                        # Official activity brief in Galician
├── enunciado.md                         # English task brief for the same activity
├── data/
│   └── sample.csv                       # Iris sample (40 rows) — targets the skills
└── doc/
    ├── backlog.md                       # Example output generated by todo-a-plan
    ├── capas.md                         # Architecture overview with file map
    ├── justificacion.md                 # Why each primitive was chosen
    └── reflexion.md                     # Required reflection on primitive choices
├── notebooks/
│   └── 01_eda_example.ipynb             # EDA notebook — targets notebooks.instructions.md
└── src/
    └── train_model.py                   # PyTorch classifier — targets python.instructions.md

How to Use This Repository

1. Read enunciado.md for the English task brief, or actividade.md if you need the official Galician wording.
2. Read doc/capas.md to understand the five layers before looking at the files.
3. Open each .github/ subfolder in layer order: instructions, prompts, skills, agents, then hooks.
4. Try the slash commands in VS Code Copilot chat: /csv-eda-basica, /comparar-primitivas, /conventional-commit, and the other packaged workflows.
5. Switch to the tutor agent in the agent picker and ask it to review a function from src/train_model.py.
6. Read doc/justificacion.md and doc/reflexion.md to connect the implementation to the primitive-selection rationale.

Prerequisites

• VS Code with the GitHub Copilot extension (agent mode enabled).
• Python 3.10+, torch, pandas, matplotlib, seaborn (for running the ML source files).
• Optional but recommended: jupyter so the notebook hook can produce a cleaned temporary copy instead of only denying the read: pip install jupyter.

Language

All repository content is in English except actividade.md, which remains in Galician as the official activity brief.