The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze and refactor Python source code, which means it ingests untrusted data from local files. This creates a surface where malicious instructions hidden in code comments or docstrings could attempt to influence the agent's behavior during the review process.
Ingestion points: The skill utilizes the Read, Glob, and Grep tools to scan Python modules and configuration files (e.g., SKILL.md instructions for code review).
Boundary markers: There are no explicit instructions for the agent to treat file content as untrusted data or to ignore embedded instructions, though standard platform safety guardrails apply.
Capability inventory: The skill possesses the Bash, Edit, and Write tools, which grant it the ability to execute shell commands and modify the local filesystem based on its interpretation of the code.
Sanitization: While the skill itself does not implement sanitization for its input, it actively teaches and provides examples of sanitization and validation for the user's own applications (e.g., the sanitize_filename and validate_user_input functions in examples/error_handling.py).

python-best-practices