The Agent Skills Directory

[COMMAND_EXECUTION]: The main instruction file SKILL.md contains a mandatory directive for the agent to execute a shell command: python scripts/generate_schematic.py "your diagram description" -o figures/output.png. This pattern is extremely vulnerable to command injection because it directly interpolates user-provided natural language ("diagram description") into a shell command without any sanitization, validation, or protective boundary markers. An attacker could provide a malicious description containing shell metacharacters to execute unauthorized commands.
[EXTERNAL_DOWNLOADS]: In references/nstc_guidelines.md, the skill instructs the user and agent to download and install LaTeX templates and packages from unverified external sources, specifically individual GitHub repositories such as github.com/L-TChen/nstc-proposal and github.com/audachang/taiwan-nstc-cm03-template. These sources are not recognized as trusted organizations or well-known services and have not been verified for safety.
[NO_CODE]: The skill's primary SKILL.md file defines the execution of scripts/generate_schematic.py as a mandatory step for generating grant figures. However, this script is missing from the provided skill files. This forces the agent to attempt to execute code that has not been supplied for security review, which is a significant safety concern.

research-grants