lov-xbti-gallery

Pass

Audited by Gen Agent Trust Hub on Sep 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes content from external repositories, including case names and descriptions, to build a gallery view. This data ingestion creates a surface for indirect prompt injection where malicious instructions could be embedded in the metadata of an external repository.
  • Ingestion points: The skill reads metadata from the cases directory of specified repositories (documented in SKILL.md, Workflow step 3).
  • Boundary markers: The execution boundary explicitly limits the scope of actions, stating that analysis and preview requests remain read-only.
  • Capability inventory: The skill uses the gh CLI for listing repository contents and a Python script (profile_store.py) for managing local configuration fields.
  • Sanitization: The workflow restricts output to specific fields (name, description, entrance URL) and prohibits the creation of new tests or submission of cases.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 7, 2026, 09:49 AM
Security Audit — agent-trust-hub — lov-xbti-gallery