The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for installing the mise tool by fetching the official installer script from https://mise.run and piping it directly to the shell (curl | sh) within CI/CD and troubleshooting contexts.
[COMMAND_EXECUTION]: The skill's primary purpose is to guide the agent in defining and executing arbitrary shell commands through the mise task runner, which are triggered by the mise run command.
[DYNAMIC_EXECUTION]: The task runner supports dynamic behavior through runtime template interpolation (using {{ ... }} syntax) and the parsing of task arguments via KDL-based usage specifications.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon configuration files that may originate from untrusted external repositories, creating a vulnerability surface.
Ingestion points: Reads and processes mise.toml, mise.local.toml, mise.lock, and executable files within the mise-tasks/ directory.
Boundary markers: The skill lacks explicit instructions or delimiters to distinguish between safe configuration and potentially malicious instructions embedded in task definitions.
Capability inventory: Provides the agent with the ability to execute arbitrary shell commands, modify the file system (via mise install), and perform network operations to download development tools.
Sanitization: No validation or sanitization mechanisms are described for the shell commands or scripts defined within the repository configuration files.

mise