The Agent Skills Directory

[COMMAND_EXECUTION]: The skill manages the lifecycle of Claude Code hooks, which involves creating shell scripts and modifying the agent's core settings.json configuration. These high-privilege operations are conducted through a safe workflow that mandates backups, syntax validation, and session restarts to ensure stability and reversibility.
[INDIRECT_PROMPT_INJECTION]: The skill converts user-defined requirements into automated hook logic, creating a surface where untrusted instructions could potentially be translated into executable scripts. This risk is effectively mitigated by the skill's operational instructions, which require manual justification for blocking hooks and emphasize a SCAN-PROPOSE-IMPLEMENT loop with user oversight.
[SAFE]: The skill includes multiple dedicated security recipes (such as security-guard-hook and skill-quarantine-hook) that provide runtime protection against secret leaks, dangerous shell commands, and unaudited third-party skills, significantly hardening the agent's execution environment.

latch