The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a localized shell script (scripts/ask-claude.sh) that invokes the claude CLI. The script implements security controls including a whitelist for model identifiers (opus, sonnet, haiku) and a restrictive case statement that only permits specific session-management flags, preventing arbitrary argument injection.- [EXTERNAL_DOWNLOADS]: The skill transmits user queries and conversation context to Anthropic's servers via the claude CLI. This is the intended behavior and the skill includes instructions to minimize the amount of context sent.- [DATA_EXFILTRATION]: By design, this skill sends data to an external provider (Anthropic). The skill instructions mandate that the agent only include necessary context and use clear delimiters (Question/Context/Instructions) to prevent unintended data leakage or instruction confusion.

ask-claude