research-add-items
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting data from external and user-controlled sources.
- Ingestion points: The workflow reads content from local
outline.yamlfiles (Step 1) and retrieves data from the internet via theWebSearchtool (Step 2B). - Boundary markers: The instructions lack specific delimiters or instructions to ignore embedded commands within the ingested research data.
- Capability inventory: The skill is granted
Writepermissions to modify local files andWebSearchfor network retrieval. - Sanitization: There is no evidence of validation or sanitization for the data retrieved from web searches or existing files before it is merged and saved to the file system.
Audit Metadata