building-chat-interfaces

Pass

Audited by Gen Agent Trust Hub on Mar 4, 2026

Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill identifies and documents patterns for an indirect prompt injection surface by interpolating untrusted external data directly into the agent's context.
  • Ingestion points: External data enters the agent context via user-provided messages (input_user_message.content) in SKILL.md and webpage metadata (URL, title, description, and headings) extracted via DOM selectors in references/chatkit-integration-patterns.md.
  • Boundary markers: The prompt construction logic in SKILL.md does not utilize delimiters (e.g., XML tags) or explicit instructions for the agent to ignore potentially malicious embedded commands within the ingested data.
  • Capability inventory: The documented backend patterns include support for tool execution (tools=[your_search_tool]) and MCP tool authentication, providing a potential path for exploitation if an injection occurs.
  • Sanitization: The provided code snippets do not include logic for sanitizing or validating external content before it is processed by the LLM.
  • [EXTERNAL_DOWNLOADS]: The skill references an external JavaScript bundle from cdn.platform.openai.com for loading the ChatKit web component. This reference targets a well-known technology provider and is part of the standard deployment strategy for the featured library.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 4, 2026, 06:37 AM