design-taste-frontend (taste-skill) — agentic threat model

Not certain from the listing — likely relies on external LLMs for design inference and code generation. It is vulnerable to prompt injection via malicious design briefs, which could trick the model into generating insecure frontend code.

Not certain from the listing — likely ingests local design briefs, existing frontend code, and design system configurations. If these source files are poisoned, the agent's output will inherit those vulnerabilities.

The agent uses a structured 'audit-first redesign flow' and 'pre-flight checklist' to orchestrate its planning and execution. Vulnerabilities include the potential bypass of the pre-flight checklist or manipulation of the audit flow via adversarial inputs.

Not certain from the listing — likely runs locally or within a developer's CI/CD pipeline. If the execution environment lacks sandboxing, the code generation process could potentially write malicious files to the host system.

Features a 'pre-flight check before shipping' and 'audit-first pass' which act as built-in evaluation and guardrail mechanisms. However, these are likely self-audits and can be gamed or bypassed by sophisticated adversarial inputs.

Not certain from the listing — as a free, open-source skill, it likely lacks built-in enterprise compliance frameworks, RBAC, or formal audit logging, relying instead on the user's environment controls.

The agent is part of a repository of ~13 sibling design skills (brandkit, brutalist, minimalist, etc.). This introduces ecosystem risks where a vulnerability or compromise in one sibling skill could laterally affect the others if they share context or execution environments.