ClawHub
Back to skill

Security audit

Meta Ads Analysis

Security checks for vulnerabilities and agentic risk

Overview

This skill is a read-only Meta Ads analysis helper with disclosed credential needs and no bundled executable code, though its activation wording should be tightened.

Install only if you intend to analyze Meta Ads account performance and are comfortable providing a read-only Meta Marketing API token and ad account ID. Prefer a token limited to ads_read, and watch for accidental activation from generic requests like 'analyze' or 'diagnose.'

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
  • 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
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (5)

Anti-Refusal Statement

High
Category
Anti-Refusal
Content
- ABO (adset-level budget): Evaluate at adset level
- Multiple ads in one adset: Evaluate at adset level, not individual ads
- Active but less than 7 days old + less than 50 results since last significant edit: Treat as still in learning phase — data is unstable
- Breakdown Effect: Meta optimizes for marginal CPA (cost of the next result), not average CPA. A segment showing higher average CPA may be protecting overall campaign efficiency. Do not judge system decisions by average CPA in breakdown reports alone.

---
Confidence
85% confidence
Finding
Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Vague Triggers

Medium
Confidence
97% confidence
Finding
The trigger phrases are broad enough to activate on common language like 'analyze' or 'diagnose' without clear Meta Ads intent. That can cause the skill to engage unexpectedly, leading to unnecessary requests for advertising credentials or unintended access to sensitive campaign data in conversations that were not meant to invoke this capability.

Description-Behavior Mismatch

Medium
Confidence
84% confidence
Finding
The manifest says this skill 'outputs structured analysis only; never provides recommendations,' but the file instructs the skill to suggest attribution windows by campaign type and later prepares session keys explicitly consumed by a recommendation skill. While some of this is routing/context, the documented behavior goes beyond purely neutral analysis into prescriptive guidance and recommendation enablement.

Intent-Code Divergence

Medium
Confidence
93% confidence
Finding
Lines L137-L138 state that landing page diagnosis is now part of 'meta-ads-recommendation → Step 4,' implying this skill no longer performs that diagnostic. However, lines L206-L208 direct this skill to write 'lp_diagnosis_general' indicating whether the problem is ad-side or landing page-side, which is a direct contradiction in documented intent.

Missing User Warnings

Low
Confidence
87% confidence
Finding
The skill documentation does not prominently warn users that it will access account-level advertising data using provided credentials. This weakens informed consent and transparency, increasing the risk that users expose or authorize sensitive business data without understanding the scope of access.

Static analysis

No suspicious patterns detected.