The Agent Skills Directory

COMMAND_EXECUTION (SAFE): The skill executes local shell commands including pwd, test, and grep to verify the project type and identify available tools (like bin/dev or background job processors). These commands are restricted to the current working directory and are used for legitimate environment discovery.
DATA_EXPOSURE (SAFE): The skill reads the project's Gemfile to check for specific dependencies. It does not access sensitive user directories (like .ssh or .aws) or environment secrets.
INDIRECT_PROMPT_INJECTION (SAFE): The skill ingests data from local files (Gemfile) and environment commands (pwd). However, the ingestion is used for boolean flag setting (HAS_JOBS, HAS_BIN_DEV) rather than open-ended prompt interpolation, significantly limiting the risk of instruction override through project files.

bootstrap