The Agent Skills Directory

[PROMPT_INJECTION]: The skill documentation includes instructions that steer agent behavior, such as 'Scientific schematics should be generated by default' and mentions of a specific persona 'Nano Banana Pro', which could be used to override default agent constraints or safety guidelines.
[PROMPT_INJECTION]: The skill presents a significant surface for indirect prompt injection as it processes untrusted data formats including PDF, Excel, and YouTube transcripts. Mandatory Evidence Chain:
Ingestion points: The markitdown.convert() method in scripts/batch_convert.py, scripts/convert_literature.py, and scripts/convert_with_ai.py processes external files and URLs.
Boundary markers: Absent. There are no instructions to the agent to treat the converted content as untrusted data or to ignore embedded instructions.
Capability inventory: The skill has access to the Bash tool and file system write operations, which could be abused if an injection is successful.
Sanitization: Absent. The scripts do not perform validation or sanitization of the content extracted from documents before it enters the agent's context.
[COMMAND_EXECUTION]: SKILL.md instructs the agent to run python scripts/generate_schematic.py, but this script is not included in the skill's file list. This could lead to the agent attempting to execute non-existent or hallucinated code.
[EXTERNAL_DOWNLOADS]: The documentation encourages downloading code and installing packages from external sources (e.g., git clone, pip install). While the referenced sources (Microsoft, Tesseract) are well-known, these operations should be performed with caution.

markitdown