Back to skill

Security audit

openai-ads

Security checks for vulnerabilities and agentic risk

Overview

The skill is a disclosed ads-management integration, but it sends reusable advertising credentials through a third-party MCP gateway and can create live budget-affecting campaigns without clear approval safeguards.

Review this carefully before installing. Use only test or least-privilege ad accounts, avoid passing long-lived provider API keys through the remote gateway, require manual approval for campaign creation, activation, budget, destination URL, and creative changes, and add privacy/legal review before using the conversion-tracking guidance with customer identifiers.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
references/client-supplied-credentials.md:9
Finding

Advertising Provider Credentials Are Transmitted to an External MCP Gateway

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (8)

Credential Access

High
Category
Privilege Escalation
Confidence
70% confidence
Finding

Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.

Content

Scanner excerpt · README.md (reported line 24)May include surrounding context.

🚀 Quick Onboarding & First-Time Setup

  1. Get your Habilis MCP Token:
  2. Configure Environment Variables:
    bash
    export HABILIS_API_KEY="hab_live_..."
    

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The README explicitly promotes autonomous campaign orchestration over a remote MCP endpoint but does not warn users that the skill can perform external network actions and potentially make live, campaign-affecting changes. In an ads-management context, this omission is risky because users may invoke the skill assuming it is informational while it can alter production advertising assets or spend.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill explicitly documents tools that can create and activate advertising campaigns, which can directly change live account state and incur real spending, yet it provides no explicit warning, confirmation requirement, or operator-safety guidance before those actions. In an autonomous agent context, this is dangerous because a user may invoke the skill expecting analysis or drafting, while the agent proceeds to perform budget-affecting actions against a live ads account.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The manifest advertises 'Autonomous AI Marketing & Ads Manager' capabilities and broad campaign management without defining approval boundaries, read-only vs write actions, or activation constraints. In an ads context, vague autonomy can normalize high-impact operations such as budget changes, campaign launches, or targeting updates without clear user awareness, increasing the risk of unintended or over-broad execution.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The manifest describes campaign orchestration, chat_cards synthesis, catalog feeds, and performance telemetry, all of which can materially affect spend, brand messaging, customer experience, and data handling, yet it provides no warning that the skill may take impactful actions. In this business/marketing context, the lack of an explicit warning or consent model makes the skill more dangerous because users may invoke it expecting analysis while it may drive externally visible or cost-incurring changes.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The guidance instructs implementers to send hashed customer identifiers and tracking tokens for conversion tracking, but it does not mention consent, notice, lawful basis, retention limits, or regional privacy compliance. Hashing identifiers with SHA-256 does not eliminate privacy risk because the data remains linkable personal data in many jurisdictions, and implementers may deploy invasive tracking without adequate user-facing disclosures.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The Portuguese section repeats instructions to transmit conversion event IDs, click tokens, and SHA-256-hashed email/phone data without warning about privacy obligations or user consent. Because this is operational implementation guidance in a marketing/ads skill, users are likely to copy it directly into production tracking flows, increasing the risk of noncompliant personal data collection and cross-system profiling.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The documentation instructs clients/agents to pass third-party provider secrets directly as tool invocation arguments, which commonly exposes those secrets to logs, traces, prompt history, debugging output, and downstream tool telemetry. In an agentic skill context, this is more dangerous because LLM tool calls may be persisted by hosts, MCP clients, or observability layers even if the gateway itself is stateless, so the 'zero-storage' claim does not eliminate exposure risk.

Content

No source excerpt is available for this finding.