# Copilot Customization Architecture

This repository keeps Copilot customizations in a layered structure under `.github/`.

## Directory Map

- `agents/` - custom agent definitions (`*.agent.md`)
- `instructions/` - always-on or file-scoped behavior rules (`*.instructions.md`)
- `prompts/` - reusable one-shot commands (`*.prompt.md`)
- `hooks/` - deterministic lifecycle hooks (`*.json`)
- `skills/` - canonical skill directories (`<skill-name>/SKILL.md`)

## Canonical Storage Rules

1. Agent definition lives at:
   - `.github/agents/<agent-name>.agent.md`
2. Skill source lives in:
   - `.github/skills/<skill-name>/SKILL.md`
4. Prompt files live in `.github/prompts/*.prompt.md`.
5. Instruction files live in `.github/instructions/*.instructions.md`.
6. Hook files live in `.github/hooks/*.json`.

## Why This Layout

- Keeps a flat, standard layout used in most repositories.
- Simplifies navigation: agents in one place, skills in one place.
- Separates deterministic automation (`hooks`) from advisory instructions.
- Keeps agent and skill lifecycles independent.

## Change Policy

- Keep only `.agent.md` files under `.github/agents/`.
- Keep only skill folders under `.github/skills/`.
- Keep descriptions trigger-rich ("Use when: ...") for discovery.
- Do not duplicate skill content in multiple places.