The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local CLI tool using absolute paths for Node.js and the renderer script. It provides specific command templates for Bash and PowerShell to ensure proper argument quoting and execution environments.
[EXTERNAL_DOWNLOADS]: The skill references a sidecar setup procedure for installing the necessary LibreOffice Kit components if they are missing. It explicitly instructs the agent to obtain user consent before initiating any downloads or installations, following best practices for software deployment.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data by reading a JSON manifest and inspecting rendered images of user-provided documents. While this is an ingestion point for external content, the risk is mitigated by the skill's specific focus on structural and visual layout analysis (e.g., checking alignment, clipping, and fonts) rather than following instructions within the document content.
[SAFE]: The skill uses unique, temporary directories with restricted permissions for rendering previews. On POSIX, it uses mktemp -d to enforce owner-only access, and on Windows, it uses GUID-based naming, ensuring that rendered images are isolated and properly cleaned up after use.

office-render