VyasaGraph — Persistent Agent Memory

• Outcome: Review
• Latest audit: May 29, 2026 at 9:29 PM UTC
• Version:
  1.2.0

Overview

VyasaGraph is a coherent local memory skill, but it asks agents to automatically save broadly defined user information into permanent memory without a clear consent or minimization gate.

Install only if you deliberately want always-on local agent memory. Before enabling it, define what categories may be remembered, avoid secrets and sensitive personal data, review or delete `memory.db` regularly, and leave `OPENAI_API_KEY` unset if stored facts should not be sent to OpenAI for embeddings.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• 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
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (3)

Vague Triggers

Medium

• Confidence: 95% confidence
• Finding: The skill explicitly instructs the agent to record substantive user information automatically and immediately in the same reply, using a broad rule of 'if the user tells you something you didn't know before, write it.' This creates a real risk of over-collection and persistence of sensitive or irrelevant data without a clear consent gate, relevance filter, or data minimization policy.

Ssd 3

Medium

• Confidence: 97% confidence
• Finding: These instructions require persistent storage of all newly learned user information in the same reply, which can cause the agent to retain personal, confidential, or unnecessary data across sessions. In an agent memory skill, that is especially dangerous because it normalizes default persistence and can turn ordinary conversations into long-term data collection without meaningful review or consent.

Ssd 3

Medium

• Confidence: 96% confidence
• Finding: The SOUL.md guidance reinforces a persistent behavioral rule that anything newly learned from the user must be written to long-term memory immediately. Because this is framed as core agent identity/instructions, it increases the chance of systematic retention of personal data and makes unsafe storage behavior persistent across interactions.

VirusTotal

65/65 vendors flagged this skill as clean.