Awesome-GUI-Agent

Maintains a systematically organized, single-file knowledge base that catalogs and cross-references academic papers, datasets, benchmarks, models, and open-source projects across five distinct GUI agent research domains (vision-language models, web navigation, mobile agents, desktop control, multimodal agents). Uses standardized entry formatting with bibliographic metadata, access badges, and temporal organization to enable rapid navigation and discovery of domain-specific resources without requiring external search infrastructure.

Integrates a custom GPT-powered agent (Awesome-Paper-Agent) that automatically generates standardized resource entries following a consistent bibliographic format with title, publication date, GitHub stars badge, arXiv badge, and website badge. The system enforces a canonical entry structure across all contributions, reducing manual formatting overhead and ensuring consistency in how papers, projects, and datasets are presented in the knowledge base.

Organizes GUI agent research across five interconnected domains (datasets/benchmarks, models/agents, surveys/literature, open-source projects, safety/security) with explicit cross-domain relationships showing how datasets inform model development, which enables practical projects, all while considering safety implications. The taxonomy structure reflects the dependency graph of GUI agent research, allowing users to trace from foundational datasets through to production implementations and safety considerations.

Classifies GUI agents into five architectural categories based on their target platform and interaction approach: vision-language models (foundation models with visual understanding), web navigation agents (browser-based task automation), mobile device agents (smartphone/tablet control), desktop control agents (OS-level application automation), and multimodal agents (cross-platform capabilities). Each category includes representative implementations and key architectural characteristics, enabling users to understand the design trade-offs and capabilities of different agent types.

Curates and organizes research on safety, security, and alignment considerations specific to GUI agents, including adversarial robustness, privacy implications of GUI automation, and risk mitigation strategies. This domain aggregates papers addressing vulnerabilities in GUI agent systems, defensive mechanisms, and best practices for safe deployment across web, mobile, and desktop platforms.

Implements a table-of-contents style navigation system that provides direct links to major resource categories (datasets/benchmarks, models/agents, surveys, open-source projects, safety/security) at the top of the README, enabling users to jump directly to relevant sections without scrolling through the entire document. This navigation infrastructure is essential for managing a large single-file knowledge base and reducing friction for users seeking specific resource types.

Tracks and organizes resources by publication date (year, venue, conference) to enable users to understand the evolution of GUI agent research over time and identify recent advances. Each resource entry includes publication metadata in parentheses, allowing users to filter by time period and understand which approaches are foundational versus cutting-edge.

Displays GitHub stars badges for open-source projects and repositories, providing a quantitative signal of community adoption and project maturity. This metric is embedded directly in resource entries, allowing users to quickly assess the popularity and active maintenance status of GUI agent implementations without visiting external sites.

Provides standardized badge links to multiple resource formats for each entry: arXiv preprints for academic papers, GitHub repositories for code, and project websites for tools and frameworks. This multi-format linking enables users to access resources in their preferred format (paper, code, or documentation) without manual searching, and supports the full research-to-implementation pipeline.