ClawHub
Back to skill

Security audit

memory-system

Security checks across malware telemetry and agentic risk

Overview

This memory skill is purpose-aligned, but it needs Review because it can automatically move private memory files into persistent PostgreSQL storage and search them with weak user controls.

Install only if you want a local long-term memory system that reads from ~/.openclaw/workspace/memory and writes to local PostgreSQL. Before using it, review the scripts, use a dedicated least-privileged database user, avoid highly sensitive memories, and add explicit review/delete controls for archived data.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (7)

subprocess module call

Medium
Category
Dangerous Code Execution
Content
"-c", f"SELECT title, content FROM memory_structured {where};"
    ]
    
    result = subprocess.run(cmd, capture_output=True, text=True)
    return result.stdout

def search_vector(query):
Confidence
95% confidence
Finding
This subprocess call executes `psql` with a SQL string built from untrusted input (`category`) via string interpolation. Although `shell=False` avoids shell injection, this is still a command-driven SQL injection path because attacker-controlled data is passed into `psql -c`, allowing arbitrary SQL execution against the database.

Lp3

Medium
Category
MCP Least Privilege
Confidence
84% confidence
Finding
The skill documentation exposes shell and file-read capabilities but does not declare permissions or boundaries for those operations. In a memory-management skill, undeclared capability to read local files and invoke shell commands increases the risk of overbroad data access and makes it harder for users or the platform to understand what sensitive resources may be touched.

Tp4

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding
The skill claims to provide long-term memory and semantic search, but the documented behavior also includes automatic archival of local files, filesystem scanning, database writes, and calls to an embedding service. That mismatch is dangerous because users may authorize a benign-seeming memory feature without realizing it can ingest broader local data and persist it externally or semi-persistently.

Missing User Warnings

Medium
Confidence
91% confidence
Finding
The documentation states that local files are automatically retained and imported into PostgreSQL after 7 days, but this is not clearly warned in the skill description or framed as requiring consent. Automatic ingestion of local content into a database can capture sensitive user data unexpectedly and expand the blast radius from local files to searchable persistent storage.

Missing User Warnings

Medium
Confidence
71% confidence
Finding
The script silently transfers archived memory contents from local markdown files into PostgreSQL, potentially changing the storage boundary for sensitive long-term memory data without explicit notice or consent. In a memory-management skill, this is more sensitive because the archived content may include user preferences, context, or other private information that operators may assume remains local.

Missing User Warnings

Medium
Confidence
89% confidence
Finding
The script transmits the user's query text to a local embeddings API without any user-facing warning, consent flow, or help text indicating that input will be disclosed to another service. In a memory system, queries may contain sensitive personal context, so silent forwarding increases privacy and data-governance risk.

Session Persistence

Medium
Category
Rogue Agent
Content
### memory_structured (结构化)
```sql
CREATE TABLE memory_structured (
    id SERIAL PRIMARY KEY,
    category VARCHAR(50),  -- goals, decisions, status, preferences, reference
    title VARCHAR(200),
Confidence
78% confidence
Finding
The skill is explicitly designed to store long-term memory, including goals, decisions, status, and preferences, in persistent tables. Persistent retention of potentially sensitive user context is risky because it creates a durable repository of personal or operational data that could be queried, misused, or exposed if access controls, minimization, and consent are insufficient.

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.