v0.2.1

agent-swift v0.2.1

Complete E2E automation tool for macOS apps via the Accessibility API. No external dependencies needed.

New since v0.1.0

• click — CGEvent mouse click by ref or x,y coordinates (SwiftUI NavigationLink support)
• fill — enter text into element by ref
• get — read element properties (text, type, role, identifier, attrs)
• find — locate element by text/role/identifier with optional chained action
• screenshot — capture app window to PNG
• is — assertions (exists/visible/enabled/focused)
• wait — poll for condition or delay with configurable timeout
• scroll — scroll direction or ref into view
• schema — command metadata JSON for AI agent integration
• press now auto-falls back to CGEvent click for SwiftUI NavigationLink
• 74 role mappings covering all 64 official macOS SDK accessibility roles
• JSON mode: --json flag, AGENT_SWIFT_JSON=1 env, or non-TTY auto-detect
• AGENT_SWIFT_HOME and AGENT_SWIFT_TIMEOUT env vars

Install

Homebrew:

brew install beastoin/tap/agent-swift

Shell script:

curl -fsSL https://raw.githubusercontent.com/beastoin/agent-swift/main/install.sh | sh

Commands (15)

doctor, connect, disconnect, status, snapshot, press, click, fill, get, find, screenshot, is, wait, scroll, schema

Binary

Universal macOS binary (arm64 + x86_64). Requires macOS 13+.
71 tests passing.

v0.1.0

agent-swift v0.1.0

CLI for AI agents to control macOS apps via the Accessibility API.

Install

Homebrew:

brew install beastoin/tap/agent-swift

Shell script:

curl -fsSL https://raw.githubusercontent.com/beastoin/agent-swift/main/install.sh | sh

Commands

• doctor — check prerequisites
• connect — connect to app by PID or bundle ID
• disconnect — end session
• status — show connection state
• snapshot — capture element tree with refs (-i for interactive only)
• press — tap element by ref

Binary

Universal macOS binary (arm64 + x86_64). Requires macOS 13+.