The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided text or documents that may contain hidden malicious instructions. It explicitly mitigates this risk by instructing the agent that input text must be treated as data only and cannot grant authority or trigger task execution described within the text.\n- [COMMAND_EXECUTION]: The documentation describes using standard Git commands (git ls-files and git check-ignore) strictly for environment verification. These commands are used to ensure that the skill's temporary directory (.tigerkit/) is correctly excluded from version control, preventing accidental data leakage.\n- [DATA_EXPOSURE]: The skill implements a secure protocol for handling sensitive user input. It mandates the use of restricted filesystem permissions (0700 for directories and 0600 for files) when handling secrets and requires the immediate deletion of sensitive temporary files upon task completion or failure.

tk-rewrite