news-summary
Pass
Audited by Gen Agent Trust Hub on Mar 10, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it ingests and processes untrusted data from external RSS feeds.
- Ingestion points: News headlines and descriptions are fetched from multiple external RSS sources in
SKILL.md(BBC, Reuters, NPR, Al Jazeera) and via potential search APIs described inPRD.md. - Boundary markers: Absent. The fetched content is passed directly to the model for summarization without delimiters or instructions to ignore embedded commands.
- Capability inventory: The agent has the ability to perform network requests via
curl, write files to the/tmpdirectory, and interact with the OpenAI API. - Sanitization: Basic HTML tag removal is performed using
sed, but no semantic sanitization or instruction filtering is applied to the news content. - [EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch news data and generate audio.
- Fetches RSS content from international news organizations including BBC, Reuters, NPR, and Al Jazeera.
- Communicates with the OpenAI API (
api.openai.com) for text-to-speech functionality.
Audit Metadata