The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads an installation script for the 'bun' runtime from the official bun.sh domain. The script's integrity is verified using a hardcoded SHA-256 checksum before execution via bash.
[COMMAND_EXECUTION]: The preamble and setup sections execute numerous shell commands to manage configurations, check for updates, and initialize environment variables. It utilizes local binaries located in the skill's own directory structure.
[DATA_EXFILTRATION]: The skill implements telemetry logging and artifact synchronization to remote repositories. These features are documented as opt-in and use interactive questions to obtain user permission before activation.
[INDIRECT_PROMPT_INJECTION]: The benchmark workflow navigates to external URLs and evaluates JavaScript to collect performance metrics, creating a surface for processing untrusted data.
Ingestion points: External URLs visited during Phase 3 and Phase 5.
Boundary markers: No specific delimiters are applied to the data retrieved from external pages.
Capability inventory: The skill utilizes Bash for execution and Write tools for saving reports to the local filesystem.
Sanitization: Data extracted from web pages via JavaScript evaluation is used for metric calculation without specific filtering for prompt instructions.

benchmark