The Agent Skills Directory

[COMMAND_EXECUTION]: Several Python scripts (e.g., run_surface_eval.py, run_unit_tests.py, init_skill.py) use the subprocess.run function to execute external commands such as git, uv, and platform-specific CLIs (codex, claude). These calls are necessary for the skill's primary function of running tests, managing repositories, and performing cross-platform evaluations. The scripts use argument lists for command construction, which is a standard mitigation against shell command injection.
[DATA_EXPOSURE]: The skill requires broad filesystem read and write permissions to manage skill libraries, generate platform projections (Codex and Claude Code), and create evaluation workspaces. The scripts use a project-root inference mechanism (searching for a categories.json marker) to scope operations to the relevant skill directory, reducing the risk of accidental access to sensitive files outside the workspace.
[EXTERNAL_DOWNLOADS]: The skill utilizes the uv package manager to execute Python unit tests. This involves downloading and managing dependencies from the official Python Package Index (PyPI), which is a well-known and standard service for software development.
[SAFE]: The skill includes defensive features such as audit_skill_inventory.py, which performs static analysis on other skills to detect risky patterns (e.g., recursive directory removal, network requests, or direct subprocess calls). This functionality improves the security posture of the skill libraries it manages.

skill-lifecycle-manager