The Agent Skills Directory

[SAFE]: The skill consists entirely of architectural documentation and design patterns for building an extensible and secure agent harness. It does not contain executable code, remote dependencies, or malicious instructions.
[COMMAND_EXECUTION]: The documentation provides an implementation example in Python using the subprocess module to demonstrate how a developer might execute hook scripts. The example is educational and correctly identifies best practices like timeouts, error handling, and output isolation.
[PROMPT_INJECTION]: The skill identifies the risk of indirect prompt injection and operator-level policy bypass. It mitigates these by providing a design that requires explicit user opt-in for hooks, enforces sandboxing, and recommends redacting sensitive data before it reaches hook processes.
[REMOTE_CODE_EXECUTION]: The architecture described allows for the execution of user-defined scripts. The skill proactively addresses the security risks of this pattern by recommending that hooks be treated as untrusted, isolated by user privilege, and loaded from verified configuration paths.

hook-system