engram-memory-protocol

Pass

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill manages a persistent memory surface where discoveries, decisions, and preferences provided by the user are stored via mem_save and recalled via mem_search. This creates an indirect prompt injection surface as retrieved memory content could influence future agent behavior. However, this is part of the core functionality for a memory-management skill.
  • Ingestion points: User-provided information about bugfixes, discoveries, and preferences captured during the session (SKILL.md).
  • Boundary markers: None identified in the instructions for tool usage.
  • Capability inventory: mem_save, mem_search, mem_context, and mem_session_summary (SKILL.md).
  • Sanitization: No explicit sanitization or validation of the content being stored is defined within the skill instructions.
  • [SAFE]: The skill utilizes specialized memory management tools to track context. These operations are described as internal bookkeeping and do not involve unauthorized network access, credential exposure, or OS-level persistence mechanisms.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 14, 2026, 09:20 PM
Security Audit — agent-trust-hub — engram-memory-protocol