The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute a local shell script pdf2png.sh when conversion is requested. The script takes user-provided file paths as arguments and processes them using standard shell quoting.
[DYNAMIC_EXECUTION]: The pdf2png.sh script contains a Python script embedded within a heredoc (PYEOF). This code is executed dynamically to access the macOS Quartz and AppKit frameworks for rendering PDF pages and stitching images.
[PROMPT_INJECTION]: The SKILL.md file includes a 'Common feedback loop' section that contains specific meta-instructions for the agent. These instructions dictate how the agent should handle user feedback, including requirements to update canonical skills and rules that invalidate previous user confirmations when certain modifications occur.
[INDIRECT_PROHIBITION_INJECTION]: The skill processes external PDF files, creating an indirect prompt injection surface.
Ingestion points: PDF file paths identified by the agent and passed to the execution script (SKILL.md).
Boundary markers: The skill does not implement specific delimiters to separate user-supplied file content from the processing instructions.
Capability inventory: The skill uses bash for script execution, python3 for document rendering, and performs file system write operations (scripts/pdf2png.sh).
Sanitization: The shell script includes a basic file extension check to ensure the input filename ends with .pdf before processing.
[EXTERNAL_DOWNLOADS]: The skill documentation specifies a dependency on the pyobjc-framework-Quartz Python package, which is required to bridge Python to the native macOS CoreGraphics APIs.

lov-pdf2png