The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/verify_render.py uses subprocess.run to launch an external web browser (Google Chrome or Chromium) for visual verification of the generated HTML. While this is the intended functionality, it involves executing shell commands to interact with system binaries. The script specifically searches for Chrome/Chromium in standard installation paths and uses the --no-sandbox flag, which reduces the browser's internal security isolation.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted PDF files, creating a surface for indirect prompt injection attacks where malicious content in a PDF could influence the agent's behavior during extraction or translation.\n
Ingestion points: scripts/extract_pdf.py extracts text and images from user-provided PDFs. The translation workflow in references/translation_workflow.md subsequently passes this content to sub-agents for processing.\n
Boundary markers: The instructions lack explicit boundary markers or instructions for the agent to ignore potentially malicious embedded commands within the PDF text blocks.\n
Capability inventory: The agent has the capability to execute scripts that perform file system writes and launch external processes (Chrome via subprocess.run).\n
Sanitization: Although scripts/build_html.py uses html.escape to sanitize text for the final HTML output, the agent is exposed to raw extracted text and rendered screenshots during the extraction and translation phases.

pdf-to-html