ClawHub
Back to skill

Security audit

Adaptive Review

Security checks across malware telemetry and agentic risk

Overview

This is a documentation-only code review skill, but its OpenClaw variant may send full code diffs to a configurable endpoint while overstating privacy and offline guarantees.

Install only if you will use trusted local or self-hosted endpoints, or you are comfortable with a remote endpoint receiving your diffs. Review diffs for secrets and proprietary content before running the OpenClaw variant, and do not rely on the stated full-privacy/offline claim unless every configured endpoint is actually local or controlled by you.

SkillSpector

By NVIDIA
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Description-Behavior Mismatch

Medium
Confidence
95% confidence
Finding
The skill markets itself as requiring no cloud API and working fully offline, but later allows arbitrary OpenAI-compatible endpoints including a cloud proxy. That mismatch can cause users to send proprietary diffs to remote services under a false privacy assumption, creating a real data disclosure risk through misleading security/privacy claims.

Intent-Code Divergence

High
Confidence
98% confidence
Finding
The statement 'Zero cloud API cost. Full privacy. Works offline.' is contradicted by later support for 'cloud proxy' and generic HTTP endpoints. In a code-review skill, this is especially dangerous because users may submit full git diffs containing secrets, internal code, or customer data while believing transmission never leaves the host.

Missing User Warnings

Medium
Confidence
94% confidence
Finding
The skill instructs users to send the review prompt and embedded git diff to an HTTP API endpoint without warning that the code may be transmitted off-host. Because the prompts include full diffs, this can expose sensitive source code, secrets in patches, or internal architecture details to remote services or intermediaries.

External Transmission

Medium
Category
Data Exfiltration
Content
### Calling the API

Use curl or your preferred HTTP client to hit the OpenAI-compatible endpoint:

```bash
curl -s "$ENDPOINT/chat/completions" \
Confidence
89% confidence
Finding
The external HTTP call itself is expected functionality for model inference, so it is not inherently malicious; however, in this skill's context it transmits review content and potentially full diffs to a configurable endpoint. That makes it a genuine security concern because the endpoint may be remote, and the surrounding documentation understates the privacy implications.

VirusTotal

VirusTotal findings are pending for this skill version.

View on VirusTotal

Static analysis

No suspicious patterns detected.