The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents and encourages the agent to interact with and execute configuration from project-local files like mise.toml and .tool-versions.
Ingestion points: Project configuration files (mise.toml, .tool-versions, .env) are read to determine tool versions and environment settings.
Boundary markers: No specific boundary markers are used in the provided instructions to differentiate between trusted system instructions and untrusted project configuration.
Capability inventory: The agent can execute arbitrary shell commands via mise run and mise exec, and modify the host environment via mise set and mise activate.
Sanitization: The documentation mentions a 'trust' mechanism (mise trust) as a security control, but the agent itself is not explicitly instructed to verify trust status before every action.
[COMMAND_EXECUTION]: The skill provides the agent with full documentation for executing arbitrary tasks and commands within the project's environment.
[PERSISTENCE]: The skill instructions and documentation guide the user/agent in modifying shell configuration files (.zshrc, .bashrc, .zprofile) to activate the tool's environment automatically in future sessions.
[EXTERNAL_DOWNLOADS]: The documentation includes instructions for downloading and installing the mise CLI directly via curl | sh from https://mise.run, which is the standard installation method for this tool.

working-with-mise