The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/pre-review.py executes the ruff linter using subprocess.run(). The execution uses a list-based argument structure instead of a shell string, which is a secure implementation that prevents command injection. The tool being executed is a well-known, legitimate linter.
[DATA_EXPOSURE]: The skill analyzes source code provided by the user. The pre-review.py script performs local file reads to extract metrics like nesting depth and function length. No network capabilities or exfiltration patterns were detected in the scripts or instructions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided source code for review, which is an inherent attack surface for indirect prompt injection.
Ingestion points: User-provided code snippets passed to the agent for analysis (SKILL.md).
Boundary markers: The instructions utilize markdown code blocks and specific step-by-step structures to separate user input from system instructions.
Capability inventory: The agent's primary capability is to provide natural language feedback and refactored code examples based on the input.
Sanitization: There is no explicit sanitization of the input code, but the agent's instructions focus strictly on architectural principles, reducing the risk of accidental instruction following.

clean-code-reviewer