pyzotero

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes bibliographic items, collections, tags, notes, and full-text document contents which can originate from shared or untrusted external sources.
  • Ingestion points: Content is loaded into the agent context through methods such as zot.items() in references/read-api.md and zot.fulltext_item() in references/full-text.md.
  • Boundary markers: There are no specific boundary delimiters or isolation instructions defined to instruct the model to ignore potential malicious prompts embedded within reference titles, abstracts, or PDF text contents.
  • Capability inventory: According to SKILL.md, the agent is granted access to high-privilege capabilities including the Bash tool.
  • Sanitization: The reference patterns do not implement text sanitization or filtering logic before rendering or processing retrieved metadata.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 01:45 AM
Security Audit — agent-trust-hub — pyzotero