AI Skills

Overview

ZeroStarter includes a .agents/skills directory containing structured instructions for AI assistants and LLM agents. These skills help AI tools like Cursor, GitHub Copilot, and other AI coding assistants perform common tasks correctly and consistently.

Skills are written in Markdown with frontmatter metadata, making them both human-readable and machine-parseable.

Available Skills

agent-browser

Location: .agents/skills/agent-browser/SKILL.md (frontmatter name: agent-browser)

A browser-automation CLI for AI agents: navigate pages, fill forms, click, screenshot, scrape data, test web apps, and even automate Electron apps. It is an external skill vendored from vercel-labs/agent-browser and tracked in skills-lock.json. The committed SKILL.md is a discovery stub; the agent loads the live workflow from the CLI itself with agent-browser skills get core.

Install: npm i -g agent-browser && agent-browser install

Turbo Build Graph

Location: .agents/skills/turbo/generate-build-graph/SKILL.md (frontmatter name: turbo-generate-build-graph)

Generates a dark-mode-friendly SVG visualization of Turborepo task dependencies.

Triggers: When the user asks to "visualize dependencies", "generate a build graph", or "update the pipeline diagram"

Workflow:

Generate the raw graph: emit DOT and render with Graphviz, removing the intermediate file. Requires Graphviz (brew install graphviz on macOS):
```
npx turbo run build --graph=graph.dot
dot -Tsvg graph.dot -o graph.svg && rm graph.dot
```
Clean up labels: strip the [root] prefix, the #build suffix, and replace the ___ROOT___ placeholder with the project name.
Style for dark mode: transparent background with GitHub-blue (#1f6feb) nodes and text.
Move to assets: save to .github/assets/graph-build.svg.

Output: a dark-mode-friendly SVG embedded in the README:

![Build Graph](.github/assets/graph-build.svg)

Skill Structure

A skill is a directory (optionally nested under a category) containing a SKILL.md:

.agents/skills/
├── agent-browser/
│   └── SKILL.md
└── turbo/
    └── generate-build-graph/
        └── SKILL.md

Externally-sourced skills like agent-browser are recorded in the root skills-lock.json with their source, sourceType, skillPath, and computedHash, so they can be re-synced and integrity-checked. There is no committed install script; the lock file is the source of record.

SKILL.md Format

---
name: skill-name
description: Brief description of what the skill does and when to use it.
---

# Skill Title

Description of the skill.

## Workflow

### 1. First Step

```bash
command to run
```

Notes about this step
Important considerations

2. Second Step

...


## Creating Custom Skills

To create a new skill:

1. Create a directory: `.agents/skills/<category>/<skill-name>/`
2. Add a `SKILL.md` file with:
   - Frontmatter with `name` and `description`
   - Clear workflow steps
   - Code examples with commands
   - Notes and edge cases

**Example Custom Skill**:

```markdown
---
name: db-migration
description: Creates and runs database migrations. Use when user asks to update database schema.
---

# Database Migration

Creates type-safe database migrations with Drizzle.

## Workflow

### 1. Create Migration

```bash
bun run db:generate

2. Review Migration

Check the generated SQL in packages/db/drizzle/

3. Apply Migration

bun run db:migrate


## Integration with AI Tools

### Cursor

Skills in `.agents/skills/` are automatically available to Cursor's AI assistant when referenced in `AGENTS.md`:

```markdown
## Skills

This project includes custom skills to assist with common tasks. Skills are located in `.agents/skills` and `.claude/skills`.
```

Today the skills live in `.agents/skills`; `.claude/skills` is the conventional Claude Code location and is picked up automatically if you add it.

### GitHub Copilot

Copilot can reference skills when you mention them in comments or prompts.

### Custom AI Assistants

The structured format makes skills easy to parse programmatically for custom AI integrations.

## Best Practices

- **Keep skills focused**: One task per skill
- **Include examples**: Show expected inputs and outputs
- **Document edge cases**: Handle errors and special situations
- **Use standard commands**: Prefer project scripts over raw commands
- **Version control**: Skills evolve with the project

On this page