The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads brand strategy and competitor data from .agents/brand.md and design-principles. If these files are populated with untrusted content, they could influence agent behavior.
[COMMAND_EXECUTION]: The skill contains a Python code snippet in references/generative-tooling.md that uses the fontTools library to outline wordmarks. This involves the agent potentially executing code to process vector paths.
[EXTERNAL_DOWNLOADS]: The skill references external design services and APIs, specifically api.quiver.ai, to generate SVG assets based on text prompts.
[DATA_EXFILTRATION]: The skill instructs the agent to send descriptive brand prompts, which may include strategic emotional targets, to the api.quiver.ai endpoint to facilitate asset generation.
[REMOTE_CODE_EXECUTION]: The skill suggests using external 'MCP' (Model Context Protocol) tools like Flora or generative surfaces like QuiverAI. While not direct shell execution, these involve offloading logic to external environments.

logo-design