The Agent Skills Directory

[SAFE]: The skill is a collection of defensive architectural patterns designed to enhance agent security. It provides tools for developers to implement 'Least Privilege' and 'Defense in Depth'.
[PROMPT_INJECTION]: The skill includes explicit logic to detect and mitigate prompt injection through semantic intent classification. It uses weighted regex patterns to flag attempts to bypass instructions or manipulate the agent's persona.
[DATA_EXFILTRATION]: Provides content filtering patterns designed to identify and block the leakage of sensitive data like API keys, secrets, and database credentials.
[COMMAND_EXECUTION]: Includes a tool-level governance decorator that allows developers to define allowlists and blocklists for dangerous tools, preventing unauthorized command execution (e.g., blocking 'shell_exec' or 'rm -rf').
[SAFE]: Code examples demonstrate secure coding practices, such as using yaml.safe_load() instead of the unsafe yaml.load() and implementing path validation to prevent directory traversal attacks in file-reading tools.

agent-governance