The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions explicitly direct the agent to install the mr CLI by piping a remote shell script directly to a shell interpreter (curl -fsSL https://raw.githubusercontent.com/JUNERDD/mr/main/install.sh | bash). This pattern allows for the immediate execution of unverified remote code on the host system.- [COMMAND_EXECUTION]: The CLI facilitates the execution of arbitrary shell commands through the MR_REQUEST_COMMAND environment variable or the mr.requestCommand Git configuration, which are processed via sh -c. Additionally, the installation process modifies the user's shell configuration files (such as ~/.zshrc or ~/.bashrc) to persist the command in the system path.- [EXTERNAL_DOWNLOADS]: The skill fetches installation scripts and binary assets from a third-party GitHub repository (JUNERDD/mr) that is not identified as a trusted source. This includes the direct execution of a downloaded script.- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface (Category 8). It processes untrusted data including Git branch names, PR titles, and environment configurations. It lacks defined boundary markers or sanitization logic when interpolating these inputs into high-privilege shell commands and repository management tasks.

mr