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

review-ui

Pass

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from user interfaces to conduct accessibility and performance reviews, creating a potential surface for indirect prompt injection.\n
  • Ingestion points: UI component sources, page text, and resource URLs are ingested as documented in references/vstack/capabilities/ui-accessibility/REFERENCE.md and references/vstack/capabilities/web-performance/references/eval-audits.md.\n
  • Boundary markers: No explicit delimiters or specific instructions for the agent to ignore embedded commands within the audited interface are defined in the workflow.\n
  • Capability inventory: The skill uses a browser evaluation tool to execute scripts for data collection as described in references/vstack/capabilities/web-performance/references/eval-audits.md.\n
  • Sanitization: The protocol requires manual inspection and sanitization of all outputs before sharing, though it does not specify sanitization for the untrusted inputs from the audited pages.\n- [DYNAMIC_EXECUTION]: The skill dynamically generates and executes JavaScript snippets to perform browser-based measurements.\n
  • Evidence: In references/vstack/capabilities/web-performance/references/eval-audits.md, the agent is instructed to write small scripts utilizing MDN-standard APIs (e.g., Resource Timing, Navigation Timing) and execute them via the browser tool's evaluate command to collect JSON-formatted metrics.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 6, 2026, 04:54 PM