Knowledge Base Ingestion
Pass
Audited by Gen Agent Trust Hub on Feb 21, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION] (LOW): Indirect Prompt Injection vulnerability surface.
- Ingestion points:
kb_ingest_textandkb_ingest_urltools (SKILL.md) allow untrusted data from users and websites into the agent's knowledge base. - Boundary markers: Absent; there are no instructions to use delimiters or ignore embedded commands within the ingested content.
- Capability inventory: Ingested content is stored in a vector database for later retrieval and agent reasoning, allowing malicious data to influence future agent actions.
- Sanitization: Absent; no sanitization or validation logic is specified for the incoming text or URL content.
- [EXTERNAL_DOWNLOADS] (LOW): The
kb_ingest_urltool performs network operations to fetch data from non-whitelisted domains, which could be used to ingest malicious content or perform basic network probing.
Audit Metadata