write-agentmemory-skill

Warn

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: MEDIUMDATA_EXFILTRATION
Full Analysis
  • [DATA_EXFILTRATION]: The skill provides instructions for the agent to access sensitive local credentials and transmit them over the network as a fallback authentication method.\n
  • Evidence: TROUBLESHOOTING.md directs the agent to read authentication secrets from local host files, specifically ~/.agentmemory/.env and ~/.agentmemory/secret.\n
  • Evidence: The document instructs the agent to use these retrieved secrets to perform authenticated network operations (REST calls) to the AGENTMEMORY_URL endpoint.\n
  • Context: The skill includes significant functional safeguards—such as forbidding the transmission of secrets to remote URLs via HTTP and limiting HTTP requests to loopback addresses—but the inherent pattern of reading sensitive credentials and transmitting them to a variable network URL constitutes an exfiltration-capable surface.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 26, 2026, 11:46 PM