The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources such as Semantic Scholar, Wikipedia, and general web searches. This creates a vulnerability where malicious instructions hidden in these sources could potentially influence the agent's behavior during the summarization process. -- Ingestion points: Data is retrieved from Semantic Scholar, Wikipedia, video transcripts, and web search results as specified in the 'Core tasks'. -- Boundary markers: The skill does not explicitly instruct the agent to ignore or delimit potentially malicious instructions within the retrieved source content. -- Capability inventory: The skill has the capability to write processed research summaries to the local filesystem within the 'literature/' directory. -- Sanitization: While the skill mentions using a 'Sanitized filename' for the output path, it does not specify any sanitization, filtering, or validation of the content retrieved from external sources before it is processed and saved.

literature-review