The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and installs shell scripts into the application configuration (settings.json), allowing for the persistent execution of code triggered by various project events like PostToolUse and PrePush.
[COMMAND_EXECUTION]: The automated installation scripts (installer.py and install-hook.sh) modify the global and local settings.json files. While intended for hook management, these scripts perform direct modifications to core application settings.
[DATA_EXFILTRATION]: A template provided by the skill (user_prompt_submit_preprocessor) is designed to log all user prompt history to a local file (.claude/prompt_log.txt). While the data remains local, this functionality creates a repository of potentially sensitive user interactions.
[PROMPT_INJECTION]: The skill processes natural language requests to generate executable hook logic, presenting a surface for indirect prompt injection. This is mitigated by a built-in validation engine, though regex-based validators can sometimes be bypassed.
[SAFE]: The skill implements a robust validation engine (validator.py) that automatically scans generated commands for destructive patterns (e.g., rm -rf, sudo), hardcoded secrets (e.g., AWS keys, JWTs), and shell syntax errors.
[SAFE]: The installer system incorporates security best practices including atomic write operations (temp file to rename) and an automated backup/rollback system that maintains the last five configuration states.

hook-factory