pdfcraft-report-bookmarks

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external PDF documents to read structures and add bookmarks, which creates a potential surface for indirect prompt injection.
  • Ingestion points: External PDF files are ingested and analyzed through the doc_open, doc_info, and bookmark_list tools as described in SKILL.md and references/craft-steps.json.
  • Boundary markers: There are no explicit boundary markers or instructions to isolate the content of the PDF from the agent's instructions during processing.
  • Capability inventory: The skill possesses capabilities to modify the file system via doc_save and execute system commands via the pdfcraft-cli tool.
  • Sanitization: There is no evidence of sanitization or validation of data extracted from the PDF before it is used to generate bookmark titles or paths.
  • [DYNAMIC_EXECUTION]: The skill uses a local Python helper script (scripts/write_script.py) to generate a JSON-based instruction set for the pdfcraft-cli tool. The execution of these dynamically generated instructions and the invocation of the helper script represent a dynamic execution pattern.
  • [COMMAND_EXECUTION]: The skill performs its tasks by invoking external commands, specifically the python interpreter to run its local script and the pdfcraft-cli tool to perform PDF operations. These are controlled subprocess calls intended for the skill's primary function.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 06:47 PM