ClawHub
Back to skill

Security audit

CogDx Calibration Audit

Security checks for vulnerabilities and agentic risk

Overview

The skill does what it says, but it can send prompt and response datasets plus wallet/payment identifiers to a paid third-party API without clear consent or sensitive-data handling guidance.

Review before installing. Use this only when you intentionally want to send calibration samples to Cerebratech/CogDx, and avoid real customer data, secrets, proprietary prompts, regulated records, or sensitive model outputs unless you have verified the provider's privacy, retention, and payment terms. Require explicit user approval before any balance check or audit request.

A.I.G

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (8)

Vague Triggers

Medium
Confidence
91% confidence
Finding
The trigger text is broad enough to activate on generic requests about confidence or uncertainty, which can cause the agent to invoke a paid external skill when the user may have intended a normal discussion or internal self-check. In this context, overbroad routing increases the chance of unnecessary data sharing and unwanted charges rather than direct code execution risk.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill instructs sending prompts, responses, confidence scores, and correctness labels to a third-party API without an explicit privacy notice, consent requirement, or guidance on sensitive data handling. That creates a real risk of leaking proprietary, personal, regulated, or otherwise confidential evaluation data to an external service.

External Transmission

Medium
Category
Data Exfiltration
Content
**1. Check if you have credit balance first:**
```bash
GET https://api.cerebratech.ai/credits?wallet=your-agent-id
```

**2. Run the audit:**
Confidence
91% confidence
Finding
This line directs the agent to contact an external endpoint with a wallet/agent identifier, which is an actual external transmission surface. While checking credits is operationally legitimate, it still exposes metadata to a third party and can contribute to tracking or unintended invocation if performed automatically.

External Transmission

Medium
Category
Data Exfiltration
Content
**2. Run the audit:**
```bash
POST https://api.cerebratech.ai/calibration_audit
Content-Type: application/json
X-PAYMENT: <x402-signature>  # or omit if using credits
X-WALLET: your-agent-id      # for credit balance payment
Confidence
96% confidence
Finding
The calibration audit call sends structured sample outputs to an external API and may also attach payment and wallet headers, creating both data exfiltration and billing risk. In this skill's context, the transmission is the core feature, but it is more dangerous because the payload can include sensitive prompts and model outputs and the skill lacks strong consent and minimization safeguards.

External Transmission

Medium
Category
Data Exfiltration
Content
# CogDx Calibration Audit API Reference

## Endpoint
`POST https://api.cerebratech.ai/calibration_audit`

## Payment Options
Confidence
90% confidence
Finding
This endpoint documents sending calibration data to an external domain, which creates a real data-exposure boundary. While external APIs are expected for this skill, the danger comes from transmitting potentially sensitive prompt/response corpora and agent identifiers off-platform without clear trust, privacy, or approval guidance.

External Transmission

Medium
Category
Data Exfiltration
Content
### Option 2: Credit balance
Include `X-WALLET: your-agent-id` header. Credits from `/feedback` submissions are deducted automatically.

Check balance: `GET https://api.cerebratech.ai/credits?wallet=your-agent-id`

## Full Request Schema
Confidence
84% confidence
Finding
The credits endpoint transmits an agent identifier in a query parameter to an external service, which can expose identifiers through logs, browser history, proxies, or monitoring systems. The risk is lower than the audit endpoint because it appears to send less content, but it still expands external data exposure.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The API schema requires submission of prompts, responses, confidence scores, and optional ground truth to an external service, but the reference does not warn that these fields may contain sensitive user, model, or proprietary data. In a calibration-audit skill, operators may naturally upload real production samples, so the omission increases the chance of unintended data exfiltration to a third party.

External Transmission

Medium
Category
Data Exfiltration
Content
- Earns $0.02–0.03 in credits
- Improves next calibration audit accuracy

Full catalog: `https://api.cerebratech.ai/catalog`
Confidence
50% confidence
Finding
Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Static analysis

No suspicious patterns detected.