skills/vllnt/skills/improve-ui/Gen Agent Trust Hub

improve-ui

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external UI components, scripts, and HTML data, creating a potential surface for indirect prompt injection attacks.
  • Ingestion points: The agent resolves platform details, screens, journeys, and consumer scripts from the project environment (SKILL.md, references/vstack/capabilities/web-performance/REFERENCE.md).
  • Boundary markers: The instructions specify that the agent must "inspect and sanitize all outputs before sharing" (references/vstack/capabilities/web-performance/references/eval-audits.md).
  • Capability inventory: The skill can modify React components, execute scripts via browser evaluation tools, and install NPM packages.
  • Sanitization: Manual review and sanitization are required for audit outputs to protect privacy.
  • [DYNAMIC_EXECUTION]: The skill utilizes runtime script generation and execution within a browser context to gather performance metrics.
  • Evidence: The document references/vstack/capabilities/web-performance/references/eval-audits.md instructs the agent to "Write each measurement as a small script... run it through the browser tool's evaluate command".
  • [EXTERNAL_DOWNLOADS]: The skill specifies a process for querying and installing packages from the NPM registry during execution.
  • Evidence: In references/vstack/capabilities/project-alignment/references/default-packages.md, the agent is directed to "query the registry and install the current release" for a set of @vllnt scoped packages.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 08:18 PM