The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to process and act upon user-controlled files such as requirements.txt, which constitutes a surface for indirect prompt injection. Maliciously crafted requirements or package metadata can execute code during the pip installation phase.
Ingestion points: Use of requirements.txt and version specifiers in SKILL.md and references/command-cookbook.md.
Boundary markers: The skill does not provide instructions to delimit or verify the content of these files before processing.
Capability inventory: The skill facilitates subprocess execution of pip install, pip install -r, and pip install -e, which can run arbitrary code via setup.py or build backends.
Sanitization: No validation or sanitization of external requirements files is instructed.
[COMMAND_EXECUTION]: The skill provides numerous instructions for the agent to execute shell commands to create virtual environments, install packages, and manage caches.
Evidence: Includes commands like python -m venv, pip install, and pip freeze across all reference files.
[EXTERNAL_DOWNLOADS]: The skill involves downloading and installing packages from remote sources.
Evidence: Refers to downloading from PyPI (https://pypi.org) and VCS repositories (git+https://github.com/org/repo.git). These are documented neutrally as they target well-known and trusted platforms.

pip