The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's primary function is to propose, configure, and debug Claude Code hooks, which execute shell scripts or prompt-based logic at specific lifecycle events. This involves modifying the user's settings.json and creating executable script files on the local filesystem.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes project-specific data (such as existing hooks and configuration files) to inform the design of new automation. Malicious instructions in these files could influence the resulting hook logic or configuration. \n
Ingestion points: Reads current hook status via /hooks and project configuration from settings.json and CLAUDE.md. \n
Boundary markers: Uses standard markdown and JSON formatting; no adversarial boundary enforcement for untrusted data is documented. \n
Capability inventory: The skill has the capability to modify system-level configuration files and create shell scripts that execute automatically. \n
Sanitization: No explicit sanitization or logic verification routines are defined for data sourced from the local project environment.
[SAFE]: The skill implements best practices for safety by including a dedicated guide for 'Secret Leak Prevention' and 'Bash Command Safety Validation'. It also explicitly warns the user about the risks of using hooks from untrusted repositories.

latch