The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill utilizes npx @basiclines/rampa to download and execute code from an external registry. The organization @basiclines is not included in the Trusted External Sources list, making this an unverifiable dependency.
COMMAND_EXECUTION (HIGH): The skill provides recipes for executing shell commands using interpolated variables, specifically <brand-color>.
Evidence (SKILL.md): rampa -C "<brand-color>" -L 98:5 --size=11 -O css --name=color
Risk: If the agent retrieves the brand color from an untrusted source (e.g., scraping a website or reading a user-provided file), an attacker could perform shell injection (e.g., by providing #000; rm -rf /) leading to unauthorized system access.
INDIRECT PROMPT INJECTION / CATEGORY 8 (HIGH): The skill presents a significant vulnerability surface due to the combination of untrusted data ingestion and shell execution capabilities.
Ingestion points: The <brand-color> parameter in SKILL.md.
Boundary markers: Absent. There are no instructions to delimit the input or ignore embedded instructions.
Capability inventory: Subprocess execution via shell command provided in the Recipe.
Sanitization: Absent. No validation logic or instruction is provided to ensure the input is a valid color hex/string before it is passed to the shell.

accessible-contrast