audit-content
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill fetches external data via the 'fetch' tool for research and processes existing articles via 'read_file', creating a surface for indirect prompt injection.
- Ingestion points: Untrusted research data from the web (fetch) and local articles (read_file).
- Boundary markers: The instructions mandate treating content as untrusted but do not provide technical delimiters (e.g., XML tags) to isolate this data.
- Capability inventory: The agent can write, create, and edit files, providing a vector for environment modification.
- Sanitization: The skill requires PII scanning and manual source verification before finalization.
- [DATA_EXFILTRATION]: While intended for research, the combination of file reading capabilities and outbound network access ('fetch') allows for potential data exfiltration if the agent is hijacked through a malicious external source.
Audit Metadata