The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to read and process workspace manifests (pixi.toml, pyproject.toml) and execute tasks defined within them.
Ingestion points: Project manifests are read to discover environment configurations and task definitions in SKILL.md and workflows/run.md.
Boundary markers: The skill does not define boundary markers or explicit warnings regarding the execution of tasks found in untrusted project manifests.
Capability inventory: The skill utilizes pixi run and pixi exec, which permit the execution of arbitrary shell commands defined in project tasks via deno_task_shell.
Sanitization: There is no evidence of sanitization or validation of task commands or interpolated MiniJinja template variables before execution.
[COMMAND_EXECUTION]: The skill's primary function is to facilitate command execution through the pixi CLI. This includes running project-defined tasks, executing arbitrary Python scripts, and managing global tool installations.
[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of dependencies from well-known registries including PyPI and Conda-forge, as well as direct downloads from Git repositories and URLs using curl as shown in documentation examples.
[PERSISTENCE]: The official documentation snapshots (e.g., references/official/global-tools.md) provide instructions for modifying shell profile files like ~/.bashrc, ~/.zshrc, and ~/.config/fish/config.fish to enable shell completions. While standard for CLI tools, this represents a persistent modification to the user's shell environment.
[DYNAMIC_EXECUTION]: The skill describes advanced task features using MiniJinja templating and the deno_task_shell runner. This allows for dynamic construction and execution of shell commands based on environment variables and user-supplied arguments at runtime.

pixi