Results for “juicebox”

Analyzes prompt injection, retrieval poisoning, memory contamination, planner drift, and tool-boundary abuse in agentic systems, mapping trust boundaries and proving exploit chains.

Analyzes runtime history to identify failures and inefficiencies, then autonomously writes improvements using a protocol-constrained evolution engine. Communicates with EvoMap Hub via a local Proxy mailbox.