The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes system utilities and compilers through the subprocess module to process documents and manage environment restrictions.
Evidence: tools/docx/scripts/office/soffice.py writes a C source file to a temporary directory and compiles it using gcc to create a socket redirection shim.
Evidence: tools/xlsx/scripts/recalc.py and tools/docx/scripts/accept_changes.py call the soffice command to interact with Office documents.
Evidence: tools/docx/scripts/equations.py calls pandoc and pip for document conversion and dependency management.
[REMOTE_CODE_EXECUTION]: The skill performs runtime software installation from external sources.
Evidence: tools/docx/scripts/equations.py automatically executes pip install to fetch the latex2mathml package from the internet if it is missing during execution.
[EXTERNAL_DOWNLOADS]: The scholarly search functionality communicates with external web services to retrieve academic metadata.
Evidence: Outbound network connections are made to api.openalex.org and api.anysearch.com in tools/paper_search/scripts/openalex_scholar.py and tools/paper_search/scripts/anysearch_academic.py.
[PROMPT_INJECTION]: The skill possesses a significant attack surface for indirect prompt injection due to its automated processing of external file content.
Ingestion points: Data is parsed from user-provided PDF documents and Excel spreadsheets in tools/pdf/scripts/extract_form_structure.py and tools/xlsx/scripts/recalc.py.
Capability inventory: The skill can execute shell commands and modify the local file system using data derived from these untrusted sources.
Sanitization: Although the skill includes an instructional 'Model Contract' framework to guide agent reasoning, it lacks programmatic sanitization or sandboxing of the raw data extracted from external documents.

math-modeling