The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several shell commands including conda, pip, and source. It performs automated installation of dependencies via pip install -r requirements.txt and pip install -e ., which results in the execution of code provided in the local project's configuration files.- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through its data ingestion process. It reads configuration data from untrusted local files such as .python-version, pyproject.toml, and setup.py. Specifically:
Ingestion points: Reads Python version and dependency requirements from .python-version, pyproject.toml, and setup.py.
Boundary markers: None present; the skill treats content from these files as trusted parameters.
Capability inventory: Uses conda create, conda env list, source activate, and pip install.
Sanitization: There is no evidence of sanitization or validation of the version strings or environment names extracted from the files before they are interpolated into shell commands. This creates a surface where a maliciously crafted version string (e.g., 3.10; rm -rf /) could lead to arbitrary command execution.

conda-setup