The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes external data, including XML responses from the arXiv API and user-provided academic papers (PDFs/BibTeX), creating a surface for indirect prompt injection.
Ingestion points: XML parsing in scripts/arxiv_query_matrix.py and file/URL processing in the academic-refchecker tool.
Boundary markers: SKILL.md contains explicit "Hard Boundaries" instructing the agent to treat tool output as evidence to inspect, not final truth, and to mark weak coverage or unsupported claims clearly.
Capability inventory: The skill has network access to fetch academic metadata and file system read access to process research papers.
Sanitization: scripts/arxiv_query_matrix.py performs basic cleaning of search terms and utilizes standard URL encoding for API requests.
[EXTERNAL_DOWNLOADS]: The skill fetches academic metadata from the official arXiv API at export.arxiv.org. This is a standard and well-known source for academic research and is handled via a local Python script.
[COMMAND_EXECUTION]: The skill executes a local Python helper script (scripts/arxiv_query_matrix.py) and specialized binaries (paper-search, academic-refchecker) located within a managed environment (${CODEX_HOME}) to perform search and reference verification tasks.

research-evidence