The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Ruby scripts and system utilities to perform backup and restore operations.
Evidence: Uses ruby lib/buttercut/backup_libraries.rb, aa extract, and unzip to process library files.
[INDIRECT_PROMPT_INJECTION]: The skill processes library names and configuration settings which could serve as a vector for shell injection if the agent interpolates untrusted metadata into command strings.
Ingestion points: Reads library names from the file system, reads configuration from libraries/settings.yaml, and accepts user input via AskUserQuestion for directory paths.
Boundary markers: None identified in the instructions to separate data from commands or to warn against untrusted filenames.
Capability inventory: The skill has the capability to execute shell commands (ruby, aa, unzip) and write to the local file system (settings.yaml).
Sanitization: There are no explicit instructions or verification steps to sanitize placeholders like <library-name> or <library> before they are used in shell commands, creating a potential command injection vulnerability surface.

backup-library