agent-research-skills: why it is not safe to use

What we found

The skill cleared our review with a score of 87 out of 100. Its research workflow carries no hidden instructions, no exfiltration paths, and no destructive commands. Two items drew advisory flags.

First, the paper_finder tool sends academic search queries to standard sources such as arXiv, CrossRef, OpenAlex, Semantic Scholar, and GitHub, and also to a non-canonical host, ai-paper-finder.info. These calls are read-only, but that one host sits outside the well-known academic APIs, so your query terms reach a party we do not recognize.

Second, the skill can read a Semantic Scholar API key from a local plaintext file (~/keys.md) to raise rate limits. This is optional, user-controlled, and stays on your machine, but it does touch a plaintext key file.

What to do instead

Use the skill if academic search fits your work. To limit where queries travel, skip the paper_finder tool and rely on the canonical sources. Store any API key in an environment variable rather than a plaintext file, and treat the key file as optional.