The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes external, untrusted PDF files that are converted into Markdown for the agent to read. This creates a potential surface where malicious instructions embedded in a PDF could influence the agent's behavior during the structuring phase.
Ingestion points: The scripts/extract_to_markdown.py script reads any PDF files located within the user-provided workspace directory.
Boundary markers: The instructions in SKILL.md do not currently specify the use of delimiters or 'ignore embedded instructions' warnings when the agent reads the generated markdown content.
Capability inventory: The agent is instructed to execute a local Python script (scripts/extract_to_markdown.py), read files from the filesystem, and use file-writing tools to generate a JSON database (parsed_invoices.json).
Sanitization: The extraction script uses the pypdf library to perform raw text extraction without additional filtering or sanitization of the content before it is presented to the agent.
Local Command Execution: The skill requires the agent to execute a Python script (extract_to_markdown.py) using the command line. This is the intended functionality of the skill to perform local processing without external API dependencies.

pdf-parsing