GitHub

Domain: Coding

The GitHub environment simulates a collaborative software-development workspace for repository management, code review, and issue-tracking workflows. It supports repository navigation, issue inspection, pull-request review, and commit-history exploration, making it suitable for evaluating agents in development-centered workflow scenarios. This environment is particularly important because software repositories combine structured metadata with unstructured code, comments, and review discussions, creating realistic opportunities for both benign collaboration and adversarial manipulation.

GUI. Representative GUI views of the simulated GitHub environment are shown in the figure, covering login, repository browsing, detailed repository inspection, issue tracking, pull-request review, and commit-history inspection.

The GitHub environment exposes a comprehensive MCP interface for repository-centered software-development workflows. As summarized in the MCP-tool table, the tool set covers environment and authentication utilities, user and organization discovery, repository lifecycle and social actions, branches, commits and file-tree operations, repository metadata, issue management, pull-request management, and activity and notification handling. These tools allow agents to browse repositories, inspect and modify project content, create commits, manage issues and pull requests, and reason over repository activity in a realistic development setting. Because software repositories combine structured metadata with code, comments, commits, and review discussions, the GitHub environment is especially useful for evaluating whether agents can safely operate over development artifacts without being manipulated by malicious instructions embedded in repository content, issue threads, or pull-request discussions.

Screenshots

Login page

Repository page

Repository detail

Issue page

Pull request page

Commits page

Simulated GitHub environment. Representative views used for repository access, issue tracking, pull-request review, and commit-history inspection in workflow-agent evaluation.