pdf-translation

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The workflow documentation describes the use of uv (a Python toolchain manager) and bun (a JavaScript runtime) to execute shell commands. Specifically, it instructs the agent or user to run local Python scripts (scripts/extract_pdf.py and scripts/split_chapters.py) and a development server in the docs/ directory.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process text from external, untrusted PDF files, which introduces a potential attack surface for indirect prompt injection.
  • Ingestion points: The skill reads data from files located in data/pdfs/ and converts them to markdown for further processing.
  • Boundary markers: The instructions do not define explicit boundary markers or "ignore instructions" warnings to prevent the agent from obeying malicious commands that might be embedded in the game rules or text being translated.
  • Capability inventory: The workflow involves file system operations (read/write), script execution via uv run, and content manipulation through an LLM translation pass.
  • Sanitization: No sanitization, validation, or filtering logic is specified to identify and neutralize malicious prompt sequences within the source documents.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 02:58 AM
Security Audit — agent-trust-hub — pdf-translation