The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is configured to download executable binary archives from an external source: https://download.compdf.com/skills/winCLI/win.zip for Windows and https://download.compdf.com/skills/macCLI/mac.zip for macOS. These domains are not recognized as trusted organizations or well-known services.
[REMOTE_CODE_EXECUTION]: The agent is instructed to download, extract, and execute the compdf_cli binary on the host machine. This represents a significant security risk as the binary content is not part of the skill's distributed code and is fetched at runtime from an external server.
[PRIVILEGE_ESCALATION]: For macOS users, the skill requires the agent to perform a chmod +x operation on the downloaded binary to grant it execution permissions, which is a step toward executing untrusted code.
[COMMAND_EXECUTION]: The skill heavily relies on executing the compdf_cli binary with numerous flags and arguments derived from user requests. This involves processing file paths and parameters through the system shell.
[INDIRECT_PROMPT_INJECTION]: The skill processes PDF documents, which are complex files that can contain hidden instructions or malicious payloads designed to influence the behavior of the AI agent.
Ingestion points: External PDF files provided by the user for operations like merge, split, or optimize.
Boundary markers: None identified; the skill does not specify delimiters or instructions to ignore content inside the PDF files.
Capability inventory: The skill has the ability to write to the file system (--output), execute shell commands, and make network requests (license activation API).
Sanitization: There is no mention of sanitizing or validating the contents of the PDF files before processing them.

pdf-editor-compdf