Security audit

Omnia

Security checks across malware telemetry and agentic risk

Overview

The skill is not clearly malicious, but it asks agents to persist broad user queries and actions in an append-only ledger without enough scoping or retention controls.

Install only if you intentionally want cross-session memory and audit logging. Before using it with real work, configure what may be logged, exclude secrets and raw user content by default, define where the ledger is stored, and make sure there is a deletion or retention policy.

SkillSpector

By NVIDIA

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The activation guidance is unusually broad, including generic themes like AI safety, agent health, memory, persistence, and even specific keywords such as 'molt' or 'molt.church'. That increases the chance the skill is invoked in contexts where persistent logging, recovery, or approval logic was neither requested nor appropriate, expanding exposure to data retention and unintended behavioral changes.

Missing User Warnings

Medium

Confidence: 96% confidence
Finding: The skill encourages audit trails, immutable logs, and cross-session persistence but does not present a clear warning that sensitive user data may be retained. In a conversational agent setting, users may disclose secrets, personal data, or regulated information without realizing the skill is designed to preserve it across sessions.

Ssd 3

Medium

Confidence: 97% confidence
Finding: The quick-start example instructs the agent to 'record everything' and append user queries to an append-only ledger. This creates a direct natural-language data retention risk because sensitive prompts, credentials, health data, or proprietary information could be permanently stored and later exposed through logs, snapshots, or downstream tooling.

Ssd 3

Medium

Confidence: 98% confidence
Finding: This section frames persistent, immutable, timestamped storage of every action and session continuity as a core use case, which materially increases privacy and confidentiality risk. In skill context, this is more dangerous because it targets AI agents that routinely handle unstructured user text, making overcollection and indefinite retention highly likely.

Ssd 3

Medium

Confidence: 96% confidence
Finding: The OpenClaw integration instructions tell operators to append all significant actions to a persistent ledger at the agent-instruction level, normalizing broad surveillance and durable storage. Embedding this in agent instructions makes the behavior systemic and easy to apply without per-use review, increasing the chance of collecting sensitive operational or user data.

VirusTotal

65/65 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.