The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates LaTeX source code by merging user observations and external literature data into a provided template, and subsequently executes shell commands (xelatex, bibtex) to compile the document. This sequence of generating a script-like file and then executing a processor on it constitutes a low-severity dynamic execution pattern.
[INDIRECT_PROMPT_INJECTION]: The skill has a large attack surface for indirect injection as it fetches content from PubMed and the general web to ground its hypotheses.
Ingestion points: Untrusted data enters via WebFetch (PubMed) and WebSearch tools as described in the search workflow.
Boundary markers: There are no instructions for the agent to use delimiters or ignore instructions within the fetched scientific text.
Capability inventory: The skill uses Write and Edit to create files and Bash to execute the LaTeX compiler.
Sanitization: No sanitization or escaping of the fetched text is performed before it is placed into the LaTeX source code.
[COMMAND_EXECUTION]: The skill utilizes the Bash tool to run the xelatex and bibtex binaries locally. This is a standard requirement for generating PDFs from LaTeX but involves the execution of external system tools on content influenced by external sources.

alterlab-hypothesis-gen