deepvista-memory

Pass

Audited by Gen Agent Trust Hub on Apr 7, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill invokes the deepvista CLI to perform data operations. Commands such as deepvista card create and deepvista card update incorporate user-supplied content into shell arguments.
  • [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface as it ingests data from automated memory and knowledge cards. * Ingestion points: deepvista memory show, deepvista memory search, deepvista card get, and deepvista card +search in SKILL.md and deepvista-vistabase/SKILL.md. * Boundary markers: Not present in the instruction set. * Capability inventory: Subprocess calls to the deepvista CLI for CRUD and search operations. * Sanitization: No explicit instructions for escaping or sanitizing retrieved data before agent processing.
  • [SAFE]: All remote references point to the vendor's official domain (app.deepvista.ai) or relate to well-known developer tools (uv). The metadata inconsistency in deepvista-vistabase/SKILL.md (labeling the card skill as deepvista-memory) appears to be a clerical error rather than a deceptive tactic.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 7, 2026, 05:25 AM