analyse-narrative

Pass

Audited by Gen Agent Trust Hub on Aug 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill documents the use of a local script, read_news.ts, executed via the bun runtime. This script is part of the agent's internal workflow for fetching and deduplicating news articles from various financial and crypto-focused sources.
  • [EXTERNAL_DOWNLOADS]: The skill identifies farside.co.uk as a primary source for Bitcoin ETF flow data. It correctly identifies that automated retrieval (via curl) may be restricted by Cloudflare and suggests appropriate technical methods (Chrome-CDP) for data extraction, which is standard practice for market data analysis.
  • [INDIRECT_PROMPT_INJECTION]: The skill inherently processes untrusted external content (news feeds). However, it mitigates potential risks by mandating a strict citation policy—requiring full URLs, dates, and verbatim quotes—and employing an analytical 'disconfirmation' framework that forces the agent to objectively evaluate information rather than blindly following instructions contained within the text.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 20, 2026, 02:21 PM
Security Audit — agent-trust-hub — analyse-narrative