The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and converts external, untrusted document formats (PDF, DOCX, XLSX, HTML, etc.) into Markdown. This creates a surface for indirect prompt injection, where malicious instructions embedded within these documents could potentially influence the agent's behavior after the conversion process.\n
Ingestion points: Files are ingested via the MarkItDown.convert() method in multiple scripts, including scripts/batch_convert.py and scripts/convert_literature.py.\n
Capability inventory: The skill has access to the Bash tool and file system write permissions, which increases the potential impact of successful injection.\n
Boundary markers: The skill does not implement boundary markers or specific instructions to the agent to ignore content found within the converted documents.\n
Sanitization: No sanitization or filtering is performed on the text content extracted from the documents.\n- [COMMAND_EXECUTION]: The skill employs subprocess execution to orchestrate internal scripts for schematic generation.\n
Evidence: scripts/generate_schematic.py uses subprocess.run to call scripts/generate_schematic_ai.py. While the implementation uses a command list to prevent basic shell injection, it establishes a pattern of dynamic script execution within the skill environment.\n- [EXTERNAL_DOWNLOADS]: The skill documentation instructs users and agents to install several external Python packages to enable its features.\n
Evidence: SKILL.md and format-specific guides recommend installing markitdown, requests, openai, and python-dotenv. While these are well-known libraries, they are external dependencies required for the skill to function.

markitdown