The Agent Skills Directory

[PROMPT_INJECTION] (SAFE): The skill is designed to handle user input for documentation purposes but includes robust proactive sanitization steps. It explicitly instructs the agent to remove path traversal sequences (..), separators (/, \), null bytes, and control characters from user-provided titles before file creation. Filenames are further restricted to a specific character set [a-z0-9-] and length (80 characters).
[COMMAND_EXECUTION] (SAFE): Uses a limited set of Bash utilities (ls, grep, sed, sort, tail) solely to calculate the next sequential ID for new files. These operations are performed on local directory paths and do not involve external input in the command string.
[DATA_EXFILTRATION] (SAFE): The skill reads local configuration and template files (e.g., .architecture/config.yml). No network tools or external data transmission capabilities are present.
[REMOTE_CODE_EXECUTION] (SAFE): No patterns for remote code downloads or dynamic execution of untrusted scripts were found. All file operations are directed to a predefined local structure.

create-adr