The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructions provide a workflow where natural language descriptions are passed directly to a bash script for figure generation, creating a potential command injection surface if user input is not properly sanitized.
Ingestion points: SKILL.md describes taking user-provided natural language descriptions (e.g., project timelines) as input.
Boundary markers: No explicit delimiters or boundary markers are used to encapsulate the user input.
Capability inventory: The skill includes allowed-tools for Bash, enabling the execution of scripts with user data.
Sanitization: The documentation does not provide instructions for sanitizing, escaping, or validating user-provided strings before they are interpolated into the shell command.
[EXTERNAL_DOWNLOADS]: The skill references the retrieval and execution of external scientific tools and templates.
Evidence: references/nstc_guidelines.md instructs the user to git clone a LaTeX repository from GitHub and install packages using the tlmgr TeX package manager.

research-grants