The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions include a command to install the software by fetching a shell script from a remote URL and piping it directly into the bash interpreter: curl -fsSL https://paperclip.gxl.ai/install.sh | bash. This pattern is considered high risk as it executes unverified remote code with user privileges.\n- [EXTERNAL_DOWNLOADS]: The documentation describes installing the Python SDK from an unversioned remote wheel file at https://paperclip.gxl.ai/paperclip.whl instead of using a standard, version-pinned package from a secure registry.\n- [PERSISTENCE]: The installation guide recommends that users modify their shell profile files, such as ~/.bashrc or ~/.zshrc, to add the tool's binary path to their environment, which is a method for maintaining the tool's availability across different shell sessions.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes extensive full-text biomedical papers and regulatory documents which could contain embedded malicious instructions.\n
Ingestion points: Data is ingested through paperclip search, grep, cat, and map commands as defined in the skill instructions.\n
Boundary markers: The skill includes explicit rules for the agent to treat all server-returned content as data and to ignore any instructions found within that content.\n
Capability inventory: The skill possesses the Bash Read Write tool, allowing it to perform local file system operations.\n
Sanitization: The security relies on the agent's adherence to natural language instructions to separate data from commands, rather than technical validation steps.\n- [COMMAND_EXECUTION]: The skill relies on executing a local binary and various shell operations, including the sourcing of .env files for environment variable management.

paperclip