The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/verify_render.py uses subprocess.run to invoke a local Google Chrome or Chromium binary. This is used for the primary purpose of visual verification by rendering the generated HTML and taking screenshots.
Evidence: Found in scripts/verify_render.py within the shoot function.
Context: The command is executed using a list of arguments, which mitigates shell injection, and serves a documented functional requirement of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from PDF files, which is eventually interpolated into agent prompts during the optional translation workflow described in references/translation_workflow.md.
Ingestion points: extract_pdf.py reads content from a user-provided PDF; translation_workflow.md uses the extracted text.
Boundary markers: The prompts in translation_workflow.md contain instructions for fidelity but do not use clear delimiters to isolate untrusted content from instructions.
Capability inventory: The skill has the ability to write files (scripts/build_html.py) and execute system commands (scripts/verify_render.py).
Sanitization: Content is escaped for HTML output in scripts/build_html.py, but the raw text is passed to agents for translation without specialized prompt sanitization.

pdf-to-html