The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided scientific papers and technical reports, which introduces a potential attack surface for indirect prompt injection if those documents contain malicious instructions designed to subvert agent behavior.
Ingestion points: scripts/extract_pdf_text.py extracts text from user-provided PDF files for analysis.
Boundary markers: static/core/principles.md defines a core principle to treat source papers as technical evidence rather than as instructions or claim language.
Capability inventory: The skill has the capability to write various file formats (DOCX, JSON, TXT, SVG, PNG) and execute internal scripts via subprocess calls.
Sanitization: No explicit sanitization or filtering is performed on the extracted text before it is processed by the AI agent.
[COMMAND_EXECUTION]: The script scripts/build_patent_package.py uses the subprocess module to coordinate the execution of other scripts within the package, including render_flowchart_svg.py and render_patent_docx.py.
[DYNAMIC_EXECUTION]: scripts/build_patent_package.py utilizes importlib.util to dynamically load and execute functions from local scripts validate_patent_draft.py and audit_claims.py to perform validation and auditing tasks.

nature-paper-to-patent