agent_types

A technical description of how device-control agents are built for different operating systems while sharing one common architecture. An agent is software that observes a device and carries out actions on it.

In plain words
What is it for?
Use it when understanding, extending, or customizing the UFO3 device-agent implementations for Windows or Linux.
Why use it?
It explains which parts are shared across platforms and which parts must fit Windows or Linux controls.

Agent

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add agents/microsoft/ufo/agent_types
Clone the repo
git clone --depth 1 https://github.com/microsoft/UFO
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 8,810 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00000 $0.08810
Opus 5 $0.00000 $0.04405
Sonnet 5 $0.00000 $0.01762
Haiku 4.5 $0.00000 $0.00881

Measured yesterday against content hash 0c0b937660c1, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

agent_types scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

documents/docs/infrastructure/agents/agent_types.md · 1,076 lines

How it starts

The opening of the file, as written. The whole thing — 1,076 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Platform-Specific Agent Implementations

This document describes how the unified three-layer Device Agent architecture is implemented across different platforms. While the core framework (State, Processor, Command layers) remains consistent, each platform implements specialized agents optimized for their native control mechanisms and hierarchies. Understanding these implementations is essential for extending UFO3 to new platforms or customizing existing agents.

Overview

UFO3's Device Agent architecture achieves cross-platform compatibility through platform-specific agent implementations that inherit from a common abstract framework. Each platform's agents implement the same BasicAgent interface while adapting the three-layer architecture to their unique execution environments:

graph TB
    subgraph "Unified Framework"
        Framework[Three-Layer Architecture<br/>State + Processor + Command]
    end
    
    subgraph "Windows Platform"
        HostAgent[HostAgent<br/>Application Selection]
        AppAgent[AppAgent<br/>Application Control]
        
        HostAgent -->|Delegates| AppAgent
    end
    
    subgraph "Linux Platform"
        LinuxAgent[LinuxAgent<br/>Shell Commands]
    end
    
    subgraph "Future Platforms"
        MacAgent[macOS Agent]
        AndroidAgent[Android Agent]
        IOSAgent[iOS Agent]
    end
    
    Framework -.Implements.-> HostAgent
    Framework -.Implements.-> AppAgent
    Framework -.Implements.-> LinuxAgent
    Framework -.Extends to.-> MacAgent
    Framework -.Extends to.-> AndroidAgent
    Framework -.Extends to.-> IOSAgent
    
    style Framework fill:#fff4e1
    style HostAgent fill:#e1f5ff
    style AppAgent fill:#e1f5ff
    style LinuxAgent fill:#c8e6c9
    style MacAgent fill:#f0f0f0
    style AndroidAgent fill:#f0f0f0
    style IOSAgent fill:#f0f0f0

Unified Framework Benefits:

  • Code Reuse: State management, strategy orchestration, and command dispatch logic shared across platforms
  • Consistent Interface: All agents implement BasicAgent interface with same lifecycle (handle, next_state, next_agent)
  • Extensibility: New platforms inherit three-layer architecture, only implementing platform-specific strategies and commands
  • Multi-Platform Coordination: HostAgent on Windows can coordinate with LinuxAgent on Linux device via Blackboard

Read the full file on GitHub · 1,076 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. yesterday First seen · 1,076 lines · 0 tokens per session scan A 0c0b937660c1

Subscribe to this mod's changes

agent_types is an agent published in the GitHub repository microsoft/UFO (9,589 stars, last pushed 8d ago), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 8,810 tokens. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.