The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the user to install the Paperclip CLI by piping a remote script to the shell: curl -fsSL https://paperclip.gxl.ai/install.sh | bash. This represents a high-risk execution pattern where remote code is executed with user-level permissions without prior verification.
[EXTERNAL_DOWNLOADS]: The skill downloads an installer script and a Python wheel from https://paperclip.gxl.ai/. While these are vendor-provided resources, they originate from a domain that is not a recognized trusted service, posing a risk of external code injection.
[CREDENTIALS_UNSAFE]: The documentation provides commands to source a .env file directly into the shell session: [ -f .env ] && { set -a; . ./.env; set +a; }. This pattern exports all variables in the file to the environment, which could inadvertently expose unrelated secrets and keys to the paperclip binary or subsequent shell commands.
[COMMAND_EXECUTION]: The skill extensively uses the Bash tool to execute paperclip CLI commands, metadata queries, and file operations on the local system.
[DATA_EXFILTRATION]: The skill documents commands such as paperclip upload and paperclip share which are designed to send local files or access permissions to external endpoints or third-party users.
[PROMPT_INJECTION]: The skill operates by ingesting and processing large volumes of full-text biomedical papers and regulatory documents from an external service. This creates an exposure to indirect prompt injection where instructions hidden in the text could attempt to influence the agent's behavior. The skill provides defensive instructions (Category 8) telling the agent to treat all service output as data and ignore embedded instructions, though no technical sanitization is enforced.

paperclip