The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses curl to fetch academic papers and associated metadata from arxiv.org. This is a well-known and established service for hosting research preprints, and the behavior is necessary for the skill's primary function of reading academic literature.
[COMMAND_EXECUTION]: Shell commands such as curl and pdftotext (from the standard poppler-utils suite) are employed to retrieve files and convert PDF content into a text format that the agent can process. These are standard system utilities for document processing and management.
[PROMPT_INJECTION]: The skill processes untrusted data from external academic papers, which creates a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through curl requests to arxiv.org and local file reads of PDF documents.
Boundary markers: The skill instructions do not explicitly define delimiters to separate the extracted paper text from the agent's internal instructions.
Capability inventory: The skill possesses capabilities for network access via curl and file extraction via pdftotext.
Sanitization: There is no specific evidence of text sanitization or filtering applied to the extracted paper content.
Mitigation: The risk is addressed by instructing the agent to adopt a skeptical researcher persona and follow a rigorous critical appraisal checklist to evaluate the validity of claims independently from the paper's own narrative.

paper-read