The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains instructions for setting up mise in GitLab CI and Docker environments that involve piping a remote script directly to a shell interpreter. This pattern is a high-risk security vulnerability as it relies on the integrity of the remote server and the transport mechanism to prevent arbitrary code execution.
Evidence: The commands curl https://mise.run | sh are explicitly provided as installation steps in the CI/CD Integration section.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to interact with and interpret project-level configuration files such as mise.toml, .tool-versions, and .env. It also supports remote configuration includes from Git and OCI registries. This creates a surface where an attacker who can commit to a repository or influence these files could inject malicious instructions or tasks that the agent might execute, especially since no explicit boundary markers or sanitization steps are defined for handling these inputs.
Ingestion points: Reads configuration from mise.toml, .tool-versions, .env, and remote URLs via the include directive.
Boundary markers: Absent.
Capability inventory: Executing shell tasks (run), managing environment variables (env, vars), and performing automated tool installations across multiple language backends.
Sanitization: Absent.

mise-expert