The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands (pbcopy, printf, osascript) through the Bash tool to perform clipboard operations. The provided templates, such as pbcopy < "<absolute-path>" and printf '%s' "<text>" | pbcopy, use double quotes which do not prevent shell expansion of subshells (e.g., $(...)) or backticks if the agent interpolates raw user input into the command string.
[DATA_EXFILTRATION]: The skill is designed to read the content of any file path provided by the user and copy it to the system clipboard. This functionality can be misused to access and expose sensitive system files, configuration files, or credentials (e.g., SSH keys, .env files) if the agent is not appropriately restricted.
[INDIRECT_PROMPT_INJECTION]: The skill includes instructions to process "previous output" or content from the conversation history. This creates an attack surface where malicious instructions embedded in previously retrieved external data (e.g., a web page summary) could trick the agent into performing unintended copy operations or accessing sensitive local paths.
Ingestion points: Conversation history and user-supplied text/paths in SKILL.md.
Boundary markers: None specified; there are no instructions for the agent to differentiate between user data and embedded instructions.
Capability inventory: Shell command execution and file reading via the Bash tool.
Sanitization: No instructions are provided for sanitizing file paths or text inputs to prevent command injection or path traversal beyond the intended scope.

copy