med-news-reporter
Pass
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a mandatory 'Iron Law' that prohibits the generation of unsourced facts. Any missing information is explicitly flagged with a placeholder [待查證], which serves as a strong internal control against hallucinations and data fabrication.
- [SAFE]: Extensive safety and ethics documentation is provided (e.g., in
media_ethics.mdandmedia_literacy.md), instructing the agent to evaluate content for defamation, privacy breaches, and legal restrictions (specifically under Taiwan law) before outputting. - [EXTERNAL_DOWNLOADS]: The reference materials list various legitimate government, financial, and fact-checking domains (e.g.,
data.gov.tw,tfc-taiwan.org.tw,mops.twse.com.tw) for manual or tool-assisted verification. These are well-known, trusted resources and are used strictly within the context of the skill's journalistic purpose. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input (transcripts and notes), but mitigates potential injection risks by requiring a multi-step 'Material Audit' where facts are explicitly listed and sourced from the provided text before drafting begins. This structural separation helps prevent the agent from mistaking instructions inside the source material for system-level commands.
Audit Metadata