narrative-news

Pass

Audited by Gen Agent Trust Hub on Aug 20, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local TypeScript scripts using the Bun runtime (e.g., .agents/skills/read-news/scripts/read_news.ts, news_store.ts) to ingest news data and maintain a local SQLite database at .cache/read-news/news.db.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external content from various news feeds. It includes mitigation measures such as a deterministic fetcher, a mandatory JSON schema for output, and a requirement for verbatim quotes and specific source URLs to prevent hallucinations or data poisoning.
  • Ingestion points: External RSS and Web feeds from 9 outlets including FT, WSJ, Bloomberg, and CoinDesk, fetched via the read_news.ts script.
  • Boundary markers: The agent is instructed to use a specific JSON schema, provide verbatim quotes, and mark paywalled content as [UNAVAILABLE].
  • Capability inventory: Local script execution and filesystem access for database management.
  • Sanitization: Relies on structural formatting and deterministic ranking (BM25 + near-dup) to manage untrusted external data.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 20, 2026, 02:21 PM
Security Audit — agent-trust-hub — narrative-news