news-summary
Pass
Audited by Gen Agent Trust Hub on Mar 2, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches news content via RSS from well-known and reputable organizations including BBC, Reuters, NPR, and Al Jazeera.
- [COMMAND_EXECUTION]: Employs standard shell utilities like grep, sed, and head to parse and format the downloaded RSS text.
- [DATA_EXFILTRATION]: Transmits summarized news text to OpenAI's speech-to-text API (audio/speech) for voice generation using the user's API key.
- [PROMPT_INJECTION]: The skill processes untrusted external data (news headlines and descriptions) which presents an indirect prompt injection surface.
- Ingestion points: External RSS feeds defined in SKILL.md.
- Boundary markers: None; summary content is passed directly to the model/API.
- Capability inventory: Network access via curl and local shell command execution.
- Sanitization: Minimal regex-based tag removal is performed, but there is no filtering for potential adversarial instructions within the news text.
Audit Metadata