deeplake-memory

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.80). The SKILL.md shows a FUSE-mounted, cloud-backed persistent memory at ~/.deeplake/memory/ that syncs with Deeplake Cloud and is shared across agents/teams (multi-agent/cloud-backed), meaning the agent will read and act on user-generated third-party files stored there and thus could be influenced by untrusted content.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill mounts a cloud-backed persistent filesystem and declares DEEPLAKE_API_URL (https://api.deeplake.ai) used at runtime to sync files that are directly visible to and can be used as prompts/memory by AI agents, so remote content from that URL can directly control agent instructions and is required for the skill to operate.