The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documents remote script installation via curl https://mise.run | sh in references/core-concepts.md and references/advanced-features.md. It also describes installing third-party plugins from arbitrary Git repositories using mise plugins install.- [PERSISTENCE]: The instructions guide users to modify shell configuration files (such as ~/.zshrc, ~/.bashrc, and ~/.config/fish/config.fish) to enable the tool's automatic activation and environment loading on every shell session.- [COMMAND_EXECUTION]: The skill covers features for automated command execution, including a built-in task runner (mise run), tool postinstall hooks, and directory-based enter and leave hooks.- [DYNAMIC_EXECUTION]: The documentation describes the Lua plugin architecture where custom tool logic is implemented in Lua scripts and executed by mise during the tool lifecycle.- [EXTERNAL_DOWNLOADS]: The skill details multiple backends (UBI, Cargo, npm, Pipx, Aqua) that download and install binaries and packages from GitHub Releases, crates.io, and other registries.- [INDIRECT_PROMPT_INJECTION]: The skill identifies a vulnerability surface where the tool automatically ingests and processes configuration data from project files to trigger environment changes and command execution.
Ingestion points: Project-level mise.toml, mise.local.toml, .env files, and legacy configuration files (.nvmrc, .tool-versions) as described in references/core-concepts.md.
Boundary markers: No explicit delimiters or warnings for the agent to ignore embedded instructions within these configuration files are provided.
Capability inventory: The tool has the capability to execute arbitrary shell commands via its task runner and lifecycle hooks as detailed in references/task-runner.md and references/advanced-features.md.
Sanitization: There is no documentation regarding the sanitization or validation of the configuration content before it is processed by the shell or the tool's execution engine.

mise