ai-interview-article

Pass

Audited by Gen Agent Trust Hub on Sep 6, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by ingesting past conversations through external tools.
  • Ingestion points: Data is retrieved via the claude-history MCP tool using the conversation_search function.
  • Boundary markers: The skill implements strict formatting rules, using quote blocks (>) to distinguish the AI interviewer from the interviewee (kimny), which provides structural separation.
  • Capability inventory: The skill is capable of generating long-form text and creating prompts for the Gemini image generation service.
  • Sanitization: While explicit programmatic sanitization is not mentioned, the skill mandates a 'Quality Gate' process and strict persona adherence which limits the impact of potentially malicious content in the history.
  • [SAFE]: The skill operates within a well-defined scope for a specific user (kimny). All external references, such as the mued.jp domain and the articles.json configuration, are consistent with the author's stated identity and the skill's utility.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 6, 2026, 07:12 AM
Security Audit — agent-trust-hub — ai-interview-article