The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (scholarly titles, abstracts, and metadata) retrieved from the OpenAlex API, which could potentially contain malicious instructions aimed at influencing the agent's reasoning.
Ingestion points: Untrusted data enters the agent context through the search_works and get_entity methods in scripts/openalex_client.py and various helper functions in scripts/query_helpers.py.
Boundary markers: The skill does not explicitly define or use delimiters (like XML tags or markdown blocks) to isolate retrieved tool output from the agent's instructions.
Capability inventory: The skill is equipped with network access to api.openalex.org and provides large amounts of text-based data for analysis and summarization.
Sanitization: There is no evidence of filtering or sanitizing the retrieved scholarly text before it is returned to the agent's context.

openalex-database