hivemind-weapon

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

• Third-party content exposure detected (high risk: 0.75). Yes: the required runtime workflow reads outsider-authored free text from the public/shared Deeplake memory files under ~/.deeplake/memory/ (e.g., summaries/ and sessions/ are “wiki summary per session” and “raw full-dialogue JSONL”), and those contents are then loaded into the agent’s LLM context; these are not authored by the operating user.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly states that Hivemind uploads and auto-recalls session content from the Deeplake cloud (https://api.deeplake.ai), which the agent will fetch at runtime and inject as memory/context that can directly influence prompts and behavior.