The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the processing and execution of Python scripts, configuration files (like pyproject.toml), and one-off code snippets which may originate from untrusted workspace environments.
Ingestion points: The agent is directed to run scripts from the filesystem (e.g., uv run <script>.py), synchronize dependencies from project files (uv sync), and execute code piped via standard input.
Boundary markers: There are no explicit instructions or delimiters provided to help the agent distinguish between legitimate code and potentially malicious instructions embedded within the files it processes.
Capability inventory: The uv CLI grants the agent the ability to execute arbitrary Python code, modify the local file system, and access the network to fetch packages from external registries.
Sanitization: The instructions do not define any validation or sanitization steps for the agent to perform on script content or package names before execution.
[DYNAMIC_EXECUTION]: The skill is centered around the dynamic execution of Python code in various forms.
Evidence: The documentation provides specific patterns for executing standalone scripts and temporary code snippets using uv run and bash heredocs.
[COMMAND_EXECUTION]: The skill provides the agent with a comprehensive set of CLI commands to manage the host's Python environment.
Evidence: Instructions include commands for project initialization, package management, and global tool installation (e.g., uv add, uv tool install, uv run).

uv-python