trigger-authoring-chat-agent

Pass

Audited by Gen Agent Trust Hub on Oct 20, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill documents the creation of an AI chat agent that processes untrusted user messages and client-provided data. This is an inherent attack surface for indirect prompt injection in chat applications.
  • Ingestion points: The run function in trigger/chat.ts accepts a messages array derived from user input and a clientData object in lifecycle hooks.
  • Boundary markers: The skill does not explicitly demonstrate the use of delimiters for user messages, but follows standard Vercel AI SDK patterns.
  • Capability inventory: The documented capabilities are limited to streaming text from an LLM provider (Anthropic) and performing defined tool calls; no arbitrary command execution or unauthorized file access patterns are present.
  • Sanitization: The skill includes a specific security recommendation to sanitize error messages in uiMessageStreamOptions.onError to prevent leaking internal stack traces or keys to the frontend.
  • [CREDENTIALS_SAFE]: The skill explicitly warns against security anti-patterns, such as minting access tokens in the browser, and demonstrates how to use server actions to keep environment secret keys secure on the server side.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 20, 2026, 06:51 AM