The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of shell commands using uvx to invoke the reducto-cli tool. It passes user-provided file paths and natural language instructions directly to the command line, which could lead to command injection if the agent does not properly sanitize these inputs before execution.
[EXTERNAL_DOWNLOADS]: The skill utilizes uvx (a tool from the uv ecosystem) to fetch and execute the reducto-cli package from a remote registry. It also includes an authentication flow (reducto login) that interacts with Reducto's external services.
[PROMPT_INJECTION]: The edit command allows for natural language instructions to modify documents (e.g., reducto edit document.pdf --instructions "..."). This represents an indirect prompt injection surface where instructions derived from untrusted data could influence the behavior of the document editing tool.
Ingestion points: User-provided strings or data extracted from files used in the --instructions flag of the edit command.
Boundary markers: None provided in the command examples to delimit instructions from potentially malicious content.
Capability inventory: The skill has the capability to execute shell commands (uvx) and modify local PDF/Office files.
Sanitization: There is no mention of sanitizing or validating the instruction string before it is passed to the shell command.

reducto-document-parsing