Agent Instructions

<!-- SPDX-License-Identifier: MIT -->
<!-- Copyright (c) PromptKit Contributors -->

---
name: agent-instructions
type: format
description: >
  Output format for persistent agent instruction files, custom agents,
  and skills. Produces composable skill files for GitHub Copilot
  (.github/instructions/*.instructions.md), custom agent definitions
  (.github/agents/*.agent.md), CLI skills (SKILL.md), and
  platform-appropriate equivalents for Claude Code and Cursor. Files
  are version-controlled and automatically loaded by agent runtimes
  including VS Code, JetBrains, GitHub.com, and the Copilot CLI.
produces: agent-instruction-file
---

# Format: Agent Instruction Files

The output MUST be one or more ready-to-commit agent instruction files,
custom agent definitions, or skill files for the specified platform(s).
For GitHub Copilot, each PromptKit protocol produces a **separate skill
file** with an `applyTo` glob so skills compose automatically — mirroring
PromptKit's own compositional architecture. When the user requests a
custom agent or CLI skill, produce the appropriate file type instead.

Do NOT produce raw prompt output with PromptKit section headers — the
content must be continuous, natural instruction prose suitable for direct
consumption by the target agent runtime.

## Output Structure

### 1. File Manifest

List every file that will be created:

| Platform | File Path Pattern | Scope |
|----------|-------------------|-------|
| GitHub Copilot (instructions) | `.github/instructions/<name>.instructions.md` | Per-skill (targeted via `applyTo`) |
| GitHub Copilot (custom agent) | `.github/agents/<name>.agent.md` | Agent persona with tools and handoffs |
| Copilot CLI (skill) | `.github/skills/<name>/SKILL.md` | Reusable workflow as slash command |
| Claude Code | `CLAUDE.md` | Project-wide |
| Cursor | `.cursorrules` | Project-wide |

For GitHub Copilot, produce **one skill file per logical concern**. The
recommended decomposition is:

| Skill file | Contents | `applyTo` example |
|------------|----------|-------------------|
| `<persona>.instructions.md` | Condensed persona identity and guardrail protocols (anti-hallucination, self-verification) | `**` |
| `<analysis-protocol>.instructions.md` | A single analysis protocol's checks and phases | Language-specific glob (e.g., `**/*.c`) |
| `<reasoning-protocol>.instructions.md` | A single reasoning protocol | `**` or task-specific glob |

State which file(s) will be produced and why.

### 2. File Content — GitHub Copilot Skill Files

Each `.instructions.md` file MUST begin with YAML frontmatter: