The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill employs a plugin-based architecture where its internal analysis modules are loaded dynamically. scripts/runner.py uses importlib.import_module to load a predefined set of internal detectors. This mechanism is constrained to the skill's own directory and does not load arbitrary or untrusted code.
[COMMAND_EXECUTION]: scripts/analyze_diff.py and scripts/run_external_tools.py use subprocess.run to interact with system utilities such as git for difference tracking and various quality tools (e.g., ruff, bandit, black). These executions are properly parameterized and restricted to well-known development tools.
[INDIRECT_PROMPT_INJECTION]: As a code review utility, the skill ingests untrusted source code. It mitigates injection risks by using static analysis (AST parsing) instead of runtime execution. Furthermore, the skill includes specialized documentation and detectors to help the AI agent identify and handle potential AI-generated artifacts or deceptive content in the analyzed code.
[EXTERNAL_DOWNLOADS]: The skill can invoke pip-audit via scripts/run_external_tools.py, which may connect to official and trusted vulnerability databases (such as OSV or PyPI) to check the status of project dependencies. This is a standard and legitimate security auditing function.

python-code-doctor