The Agent Skills Directory

Terminal Session Management: The skill utilizes wezterm cli spawn to create interactive terminal panes. This is used to run independent Claude Code sessions, which is a core requirement for its benchmarking and evaluation functionality.
External Resource Retrieval: It executes npx add-plugin to download and install a plugin directly from the Vercel Labs GitHub repository. This is a standard procedure within the development environment to extend the agent's capabilities for testing.
Localized File System Access: The skill interacts with specific local directories such as ~/.claude/debug/ and ~/dev/vercel-plugin-testing/. This access is necessary for organizing test projects and analyzing the debug logs generated during agent evaluations.
Data Analysis Surface: To verify agent performance, the skill reads and parses content from debug logs. Since these logs contain output generated by other AI sessions, this serves as the primary ingestion point for the evaluation process. This pattern is essential for the skill's goal of monitoring and verifying multi-system builds.

benchmark-agents