The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous templates for shell commands to be executed via hooks, including file protection scripts, code formatters (npx prettier, ruff), and desktop notifications (osascript, notify-send). These are standard development automation tasks within the context of the platform.
[DYNAMIC_EXECUTION]: The skill documents the platform's hook architecture which supports executing shell commands (type: command), LLM evaluations (type: prompt), and subagents (type: agent) at runtime. This dynamic execution is triggered by lifecycle events like tool calls or session starts.
[INDIRECT_PROMPT_INJECTION]: As the skill documents hooks that ingest external data (tool inputs/outputs) and inject it back into the agent context (additionalContext), it identifies a potential attack surface. However, it explicitly includes a 'Security Considerations' section advising users to validate inputs using tools like jq and to avoid processing sensitive files, effectively providing remediation guidance.
[SAFE]: The skill includes security-positive templates, such as a 'PreToolUse Validator' to block destructive commands like rm -rf and a 'Protected File Blocker' to prevent modifications to sensitive files like .env and .git/ directories.

meta-hook-creator