The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill processes untrusted PDF documents which may contain embedded instructions intended to influence agent behavior.
Ingestion points: The skill accepts one or more PDF files for analysis as described in SKILL.md.
Boundary markers: SKILL.md includes explicit defensive instructions to treat PDF content as untrusted data and ignore any embedded commands attempting to alter the workflow.
Capability inventory: The skill performs file system operations including reading PDFs, writing JSON diagnostics/manifests, saving PNG crops, and creating ZIP archives.
Sanitization: All file operations are managed by a specialized Python script (pdf_image_extractor.py) that enforces strict security controls, including path traversal prevention, symlink traversal blocking, and localized execution within an approved directory.
[Command Execution]: The workflow relies on executing a local Python helper script to perform rendering, cropping, and packaging tasks.
Evidence: SKILL.md workflow steps invoke python scripts/pdf_image_extractor.py with specific arguments.
Context: This is an expected pattern for complex document processing. The script is designed with security-first principles, ensuring that command arguments (like document IDs and paths) are validated and sanitized before use.

semantic-pdf-image-extractor