Tools Overview

Every SkyKoi agent has access to a powerful tool system that goes far beyond simple chat. These tools give your agent real-world capabilities on its dedicated EC2 instance.

File Operations

• •read — Read text files and images. Supports offset/limit for large files.
• •write — Create or overwrite files. Automatically creates parent directories.
• •edit — Precise text replacement (find-and-replace). The old text must match exactly.

Shell Commands

• •exec — Execute shell commands with full access (exec.security: "full"). Supports timeout, working directory, environment variables, background execution, and PTY mode for interactive CLIs.
• •process — Manage running exec sessions: list, poll, log, write stdin, send keys, kill.

The agent runs commands directly on the EC2 instance with exec.host: "gateway". This means the agent has access to everything installed on the server, including Node.js, AWS CLI, Docker, git, and any other tools.

Web & Search

• •browser — Full browser automation: navigate, snapshot, screenshot, click, type, evaluate JavaScript, handle dialogs, upload files
• •web_search — Search the web via Brave Search API with region and language filtering
• •web_fetch — Fetch and extract readable content from URLs (HTML → markdown/text)

Media & Communication

• •image — Analyze images with vision models
• •tts — Convert text to speech via ElevenLabs
• •message — Send messages across all connected channels with full formatting support
• •voice_call — Make and manage phone calls with voice conversation support

Device & Automation

• •nodes — Control paired iOS/Android/macOS devices: camera, screen recording, location, notifications
• •canvas — Present/evaluate/snapshot web UIs on node displays
• •cron — Schedule recurring tasks with cron expressions (runs in isolated sessions)

Agent Management

• •Sub-agent spawning — Spawn isolated sub-agents for complex tasks, parallel work, or background processing
• •Session management — List, read history, send messages to, and spawn new sessions
• •Memory — Read/write workspace memory files for persistent context across sessions

Tool Policy

Tool access can be configured per agent. You can allow/deny specific tools and set per-agent access profiles:

{
  "tools": {
    "allow": ["read", "write", "exec", "web_search", "web_fetch", "message"],
    "deny": ["browser", "nodes", "canvas"]
  }
}