Plugins

UltraSushiTron plugins are WASM modules that extend agent capabilities. Each plugin declares a type and a set of capabilities that are validated at install time and enforced at runtime.

Plugin Types

Every plugin declares one of four types via the plugin_type field. The type determines which capabilities are valid and how the plugin integrates with the agent runtime.

If plugin_type is omitted, the plugin defaults to tool.

Manifest Format

Each plugin ships a plugin.json manifest alongside its WASM binary. The manifest declares metadata, capabilities, and the entry point function.

Example: Tool Plugin

{
  "name": "hello-plugin",
  "version": "0.1.0",
  "description": "Example plugin demonstrating UltraSushiTron plugin system",
  "capabilities": [
    "logging",
    {
      "type": "secret_request",
      "secret_types": ["api_key", "token"]
    }
  ],
  "entry_point": "execute",
  "author": "UltraSushiTron Team"
}

Example: Browser Agent Plugin

{
  "name": "browser-agent",
  "version": "0.1.0",
  "description": "Example plugin demonstrating browser automation capabilities",
  "capabilities": [
    "logging",
    {
      "type": "browser",
      "allowed_domains": ["example.com", "httpbin.org"],
      "allowed_actions": ["navigate", "get_text", "click", "wait_for_selector"],
      "headless_only": true
    }
  ],
  "entry_point": "execute",
  "author": "UltraSushiTron Team"
}

Capability System

Capabilities are the security boundary between plugins and the host system. Each capability must be declared in the manifest and approved by the user at install time. The runtime enforces these grants — undeclared capabilities are blocked.

Type-Specific Restrictions

Not all capabilities are valid for all plugin types. The runtime rejects manifests that declare capabilities outside their type's allowed set.

Marketplace Catalog

The UltraSushiTron marketplace ships 23 verified plugins across 7 categories. Browse and install them from the Plugin Marketplace.

Security

Cloud Providers

CI/CD

AI Integrations

Monitoring

Chat Connectors

Developer Tools

WASM Sandbox

All plugins execute inside a WASM sandbox with strict isolation:

• No direct syscalls — plugins communicate with the host exclusively through declared capability imports
• Memory isolation — each plugin instance gets its own linear memory
• Deterministic execution — no access to clocks, random, or non-deterministic APIs unless explicitly granted
• Resource limits — execution time and memory consumption are bounded by the host

Marketplace Security Pipeline

Plugins published to the marketplace undergo automated security analysis before approval:

1. WASM binary analysis — imports, exports, memory pages, WASI usage, suspicious imports
2. Capability audit — declared capabilities compared against detected usage; mismatches flagged
3. LLM review — AI-assisted code review with approve/flag/reject recommendation
4. Risk scoring — composite score determining automatic approval or manual review

Admin Controls

Installed plugins can be managed from the mobile admin interface (Plugins tab) or via the Admin API:

• List plugins — view all installed plugins with type, version, and enabled state
• Toggle plugins — enable or disable individual plugins at runtime
• View capabilities — inspect the granted capabilities for each plugin