The Agent Skills Directory

[OBFUSCATION]: The skill encourages the use of a shortened URL (https://bit.ly/SPFinit) for remote code execution. Shortened URLs hide the ultimate destination of the script, preventing users and security tools from verifying the source or integrity of the code before it is executed.
[REMOTE_CODE_EXECUTION]: The skill provides specific patterns for downloading and executing SAS code directly from remote GitHub repositories and shortened URLs. Using the %include statement with a remote URL allows for the execution of arbitrary scripts in the SAS session.
[DYNAMIC_EXECUTION]: The SAS Packages Framework described in the skill is designed to download, unzip, and dynamically compile code into the SAS environment at runtime. This process of loading external macros into the session represents a dynamic execution pattern.
[INDIRECT_PROMPT_INJECTION]: The skill describes a system for ingesting and executing code from external archives (SASPAC). This constitutes a vulnerability surface where the agent could be manipulated into processing malicious packages.
Ingestion points: The %installPackage macro fetches external zip files from remote repositories into a local directory.
Boundary markers: The instructions do not describe any integrity verification (such as checksums) or boundary markers to prevent the execution of malicious embedded instructions within the packages.
Capability inventory: The framework possesses network access for downloads, filesystem access for unzipping, and the ability to execute code via compilation into the SAS session.
Sanitization: No sanitization or validation of the downloaded package content is documented in the usage instructions.

sas-packages