Agentify

Getting Started

Build a foundation for AI-assisted development in your project.

What you’ll get:

  • Layer 1 (immediate): Safe agent rules that work right away
  • Layer 2 (optional): Project context for smarter responses
  • Layer 3 (optional): Specialized skills and workflows

Prerequisites

  • A code project (any language, any size)
  • An AI coding assistant (any agent, including but not limited to GitHub Copilot, Claude Code, Codex, Cursor, Windsurf, and others)

Installation Methods

# Clone the repository
git clone https://github.com/YOUR_USERNAME/agentify.git

# Copy framework to your project
cp -r agentify/framework your-project/.agentify

# Clean up
rm -rf agentify

Method 2: Manual File Copy

Copy these files to your-project/.agentify/:

  • AGENTS.template.md
  • setup.prompt.md
  • questionnaires/ folder
  • bootstrap-prompts/ folder
  • templates/ folder

Running Setup

You have two options: agent-driven (recommended) or manual.

Step 1: Start your AI agent

Open your project in your preferred AI tool.

Step 2: Run the setup prompt

Tell your agent:

Run .agentify/setup.prompt.md

Or:

Read .agentify/setup.prompt.md and follow the instructions

Step 3: Answer questions

The agent will ask:

  1. Project name — e.g., “MyApp Backend”
  2. One-sentence description — e.g., “REST API for mobile application”
  3. AI tool — Which tool are you using? (Copilot/Cursor/Claude/Other; choose Other for Codex, Windsurf, and unlisted tools)

KB files are created or updated later via .agentify/questionnaires/kb-builder.md or .agentify/bootstrap-prompts/repo-scan.prompt.md.

Step 4: Verify output

Agent creates:

File Purpose
AGENTS.md Main rules file
docs/ Knowledge Base folder
Skills folder For custom skills
Prompts folder For workflows
Tool config .cursorrules, copilot-instructions.md, etc.

Option B: Manual Setup (if agent fails)

If agent-driven setup doesn’t work, you can set up manually:

Step 1: Copy template

cp .agentify/AGENTS.template.md AGENTS.md

Step 2: Fill required values in AGENTS.md

  • `` → Your project name (e.g., “MyApp Backend”)
  • `` → One sentence (e.g., “REST API for mobile application”)
  • Keep constraints as a list in docs/constraints.md

Step 3: Create folder structure

mkdir -p docs
# For GitHub Copilot:
mkdir -p .github/skills .github/prompts
# For Cursor: mkdir -p .cursor/skills .cursor/prompts
# For Claude: mkdir -p .claude/skills .claude/prompts

Step 4: Add tool config (optional)

Use the config format that matches your tool:

Tool Config
GitHub Copilot .github/copilot-instructions.md with: Read [AGENTS.md](../AGENTS.md) for agent instructions.
Cursor .cursorrules with: Read AGENTS.md in project root for instructions.
Claude Code CLAUDE.md with: Read AGENTS.md for agent instructions.
Codex / Windsurf / Other Point your tool to AGENTS.md in project root

Include workflow activation order in tool config:

  • Before generic actions, check prompts folder workflow **Trigger** phrases against user request
  • If matched, read and execute that workflow first
  • Then load relevant skills

After Setup — Layer 1 Complete

You now have Layer 1: Safe defaults. Your agent will follow the rules in AGENTS.md.

Try it:

Review my recent changes

or

Add a new endpoint for user preferences

The agent should now:

  • Ask before making breaking changes
  • Admit when it doesn’t know something
  • Look for existing patterns before writing new code

Growing Your Foundation

When to add Layer 2 (Knowledge Base)

Signs you need it:

  • Agent doesn’t understand your domain terms
  • Agent makes wrong assumptions about architecture
  • You keep explaining the same context

How to add:

Run .agentify/questionnaires/kb-builder.md

This creates or updates:

  • docs/project_context.md — Project overview
  • docs/glossary.md — Project terminology
  • docs/domain.md — Business entities and rules
  • docs/architecture.md — System overview
  • docs/constraints.md — Project rules

It can also suggest project-specific rules for AGENTS.md and apply them only after your approval.

Ongoing maintenance default: after architecture/domain/constraints/contract changes, check KB impact and update affected files in docs/. The default code-review workflow includes a required KB Impact result (none or updated files).

When to add Layer 3 (Skills & Workflows)

Signs you need it:

  • You want consistent code style checks
  • You have a specific review process
  • Agent needs domain expertise (API design, security, etc.)

How to add:

Run .agentify/questionnaires/skills-builder.md

This flow:

  • Reads KB context from docs/project_context.md, docs/domain.md, docs/constraints.md, docs/architecture.md, and docs/glossary.md (if present)
  • Combines KB with your answers to draft skills/workflows
  • Uses create-or-update behavior for existing files
  • Requires explicit confirmation before writes: yes / edit / no
  • Uses naming: *.skill.md for skills, *.prompt.md for workflows

Alternative: Auto-scan Repository

Let the agent analyze your codebase and draft KB:

Run .agentify/bootstrap-prompts/repo-scan.prompt.md

Folder Structure Explained

After setup + KB build (example shows GitHub Copilot paths):

your-project/
├── AGENTS.md              # Layer 1: Safe defaults
│                          # Agent reads this FIRST
│
├── docs/                  # Layer 2: Knowledge Base
│   ├── project_context.md # "What is this project?"
│   ├── glossary.md        # "What does X mean?"
│   ├── domain.md          # "What are the business rules?"
│   ├── architecture.md    # "How does this system work?"
│   └── constraints.md     # "What rules must be followed?"
│
├── .github/               # Layer 3: Skills & Workflows
│   ├── skills/            #   (Copilot uses .github/)
│   │   └── code-quality.skill.md
│   └── prompts/
│       └── code-review.prompt.md
│
└── .agentify/             # Framework source
                           # Don't modify - use for re-setup

Tool-specific paths for Layer 3:

Tool Skills & Workflows Location
GitHub Copilot .github/skills/, .github/prompts/
Cursor .cursor/skills/, .cursor/prompts/
Claude Code .claude/skills/, .claude/prompts/
Other Your choice

For Codex and Windsurf, choose Other during setup.

Customization

Edit Rules

Open AGENTS.md and edit the “Project-Specific Rules” section:

### Project-Specific Rules

- All APIs must be backward compatible
- Use repository pattern for data access
- Log all external API calls

Add Skills

Copy .agentify/templates/skills/skill.template.md to your skills folder and save as [name].skill.md.

Add Workflows

Copy .agentify/templates/workflows/workflow.template.md to your prompts folder and save as [name].prompt.md.

Troubleshooting

Agent doesn’t find setup.prompt.md

Make sure the path is correct:

  • .agentify/setup.prompt.md (with dot prefix)
  • Check file exists in your project root

Agent creates wrong paths

Run setup again and choose the correct AI tool when asked.

Rules not being followed

  1. Check AGENTS.md exists in project root
  2. For Copilot: Verify .github/copilot-instructions.md points to it
  3. For Cursor: Verify .cursorrules mentions it
  4. Verify tool config includes workflow trigger checking before default behavior
  5. Try explicitly asking: “Read AGENTS.md, check workflow triggers, then proceed”

Workflow trigger not activating

  1. Check workflow file is in prompts folder and uses *.prompt.md naming
  2. Confirm workflow has a **Trigger**: "phrase 1", "phrase 2" line near top
  3. Test exact trigger phrases first (for default code review: “Review my changes”, “Code review”, “Review current branch”)
  4. If exact phrases work but variants fail, add explicit phrase variants to **Trigger**

Next Steps

Growth path:

  1. Use Layer 1 for a few days/weeks
  2. When agent lacks context → add Layer 2 (KB)
  3. When you want specialized behaviors → add Layer 3 (Skills/Workflows)

Learn more: