compact-boundary-session-distillation

Pass

Audited by Gen Agent Trust Hub on Jun 27, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill provides logic for distilling session memory and does not contain any executable scripts, remote downloads, or network-connected tools.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data to generate context summaries. It correctly mitigates the risk of indirect prompt injection by wrapping external content in explicit XML-style tags (<session_memory> and <user_messages>) and by using a specific utility (getMessagesAfterCompactBoundary) to strictly limit the scope of the ingested transcript to recent messages.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 27, 2026, 08:02 PM
Security Audit — agent-trust-hub — compact-boundary-session-distillation