pdfcraft-ocr-scans

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references external binaries and model files for the PdfCraft engine.
  • Evidence: references/craft-command-evidence.json points to a portable Windows executable hosted at github.com/storytold/pdfcraft.
  • Evidence: references/ocr-models.json contains URLs for .rten models hosted on ocrs-models.s3-accelerate.amazonaws.com.
  • Note: All downloads are accompanied by SHA256 hashes for integrity verification.
  • [DYNAMIC_EXECUTION]: The skill uses a CLI runner to execute task sequences defined in JSON format.
  • Evidence: references/craft-methods.md describes executing pdfcraft-cli run --script steps.json, where steps.json defines a sequence of tool calls.
  • [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by ingesting and processing untrusted document content.
  • Ingestion points: Document data enters via doc_open and ocr_recognize as defined in references/craft-command-evidence.json.
  • Boundary markers: The skill lacks explicit boundary markers or instructions to ignore embedded commands within the OCR'd text during extraction.
  • Capability inventory: The skill can perform file system writes (doc_save) and execute local CLI commands.
  • Sanitization: There is no evidence of sanitization or filtering of the text extracted from the PDF before it is returned to the agent context.
  • [COMMAND_EXECUTION]: The instructions involve executing local CLI tools and managing node packages.
  • Evidence: SKILL.md suggests installing a related skill via npx skills add full-aigc-skills/pdfcraft-skills.
  • Evidence: references/craft-methods.md details usage of pdfcraft-cli for processing documents.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 06:47 PM