The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to perform extensive web searches and fetch data from external sources like arXiv and general websites to generate reports. It lacks explicit instructions to treat this external content as untrusted or to use boundary markers (delimiters) when processing it. This creates a surface where malicious content on a web page could potentially influence the agent's output or behavior during the synthesis phase.
Ingestion points: The skill uses WebSearch, WebFetch, and the arxiv.org API (via curl) to retrieve data from the public internet (specified in SKILL.md).
Boundary markers: There are no specific instructions in the methodology to wrap retrieved content in delimiters or to ignore embedded instructions within the fetched data.
Capability inventory: The skill possesses file-writing capabilities through its scripts/md_to_pdf.py script and instructs the execution of shell commands for package installation and document conversion.
Sanitization: No sanitization or validation logic is defined for the content fetched from the web before it is passed to the LLM for report generation.
[COMMAND_EXECUTION]: The skill's documentation (SKILL.md) instructs the agent or user to execute shell commands, including pip install for dependencies and running a local Python script (scripts/md_to_pdf.py) to generate PDF outputs. This is part of the core functionality but involves the execution of code on the host system.
[EXTERNAL_DOWNLOADS]: The skill fetches data from the well-known academic service arXiv (export.arxiv.org) using curl. This is a legitimate and expected operation for a research tool.

zj-systematic-research