MemOS Dreaming

• Outcome: Review
• Latest audit: May 29, 2026 at 9:34 PM UTC
• Version:
  1.0.0

Overview

This skill is a mostly coherent local memory-consolidation tool, but it can automatically rewrite and delete persistent memory content with weaker user controls than the risk warrants.

Install only if you want automated local memory consolidation. Run both scripts without --apply first, inspect DREAMS.md and AUDIT.md, and verify any cron jobs before enabling them. Be especially cautious with memos_dreaming_audit.py --apply because it can delete orphan memory entries, not just obvious noise; keep MEMORY.md under backup or version control.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access

Findings (4)

Intent-Code Divergence

High

• Confidence: 98% confidence
• Finding: The auto-clean function claims to remove only high-confidence noise, but it also deletes orphan entries unconditionally. In a memory-management skill, this can silently remove legitimate user data based on structure alone, creating integrity and availability risk because valid notes may be erased during routine maintenance.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The skill explicitly documents scheduled cron jobs that write to MEMORY.md, overwrite DREAMS.md, and may auto-clean content during audit apply mode, but it does not present a clear safety warning about these autonomous file modifications. In an agent environment, undocumented automatic writes and overwrites increase the risk of unintended data loss, confusing state changes, and silent persistence of derived content, especially because the jobs run unattended on a schedule.

Missing User Warnings

Medium

• Confidence: 90% confidence
• Finding: The script automatically writes consolidated content from MemOS and daily memory logs into DREAMS.md, which may contain sensitive personal or operational information, without any explicit consent gate, warning, or content redaction. In this context, the skill processes memory data from local workspace files and a SQLite database, so silent persistence of extracted summaries increases the risk of unintended disclosure to other tools, users, or sync processes.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: When run with --apply, the script writes changes to MEMORY.md immediately after analysis without an interactive confirmation step or dry-run approval. In this skill context, the file appears to be a persistent user memory store, so unintended invocation or misclassification can lead to destructive state changes that are hard to notice until after data is lost.

VirusTotal

64/64 vendors flagged this skill as clean.