sdlc-convert-doc

Pass

Audited by Gen Agent Trust Hub on Oct 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/convert_to_markdown.py executes external command-line utilities (markitdown and marker_single) using subprocess.run. The commands are constructed using a list format, which avoids shell injection vulnerabilities, and the executables are identified via shutil.which or user-defined arguments.
  • [INDIRECT_PROMPT_INJECTION]: This skill converts untrusted external data (various document formats) into Markdown text for agent consumption, creating an indirect prompt injection attack surface.
  • Ingestion points: Local files passed to the input argument in scripts/convert_to_markdown.py.
  • Boundary markers: The script does not wrap the converted output in protective delimiters or provide specific 'ignore' instructions to the agent when it reads the resulting Markdown.
  • Capability inventory: The skill can execute shell commands via subprocess.run (to run converters) and perform file system read/write operations.
  • Sanitization: No content sanitization or instruction-filtering is performed on the extracted text before it is returned to the agent.
  • [EXTERNAL_DOWNLOADS]: The skill requires and provides instructions for installing markitdown (from Microsoft) and marker-pdf (from Datalab) via pip. As noted in the documentation, marker-pdf may download machine learning models from remote repositories during its initial execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 15, 2026, 03:36 AM