Back to skill

Security audit

ads-case-analyzer

Security checks for vulnerabilities and agentic risk

Overview

This looks like a legitimate ad troubleshooting skill, but it needs review because it handles sensitive internal ad data with broad SQL templates and references privileged impersonation tooling without clear safeguards.

Install only for authorized analysts who are allowed to access the referenced internal ad systems. Before use, require explicit account or campaign identifiers, validate dates and IDs, use read-only scoped database roles, avoid free-form SQL predicates, and do not use impersonation tooling unless a separate approved and audited process requires it.

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 (3)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:119
Finding

SQL Injection Through Unsanitized Brand Account and Date Placeholders

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
references/adtrace_funnel.md:152
Finding

SQL Injection Through Dynamically Constructed Funnel Query Identifiers

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
references/sql_templates.md:408
Finding

Arbitrary SQL Predicate Injection Through the id_filter Template

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The documentation directs users to an impersonation/login capability ('附身登录') to inspect creative review status. Impersonation tooling is highly sensitive and not necessary for the stated purpose of pulling performance data and generating a root-cause report; if followed, it could enable unauthorized account access, privilege misuse, and access to nonessential user or advertiser data. The skill context makes this especially dangerous because it normalizes privileged access as part of standard troubleshooting guidance.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger phrases are broad, natural-language support requests that overlap with common user asks, which can cause the skill to activate in situations where the user did not intend this specific workflow. Because the skill accesses internal advertiser and bidding datasets, unintended activation can lead to unnecessary retrieval or analysis of sensitive business data. The context increases risk because the skill is powerful and data-intensive, not a harmless helper.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill description does not warn users that it accesses potentially sensitive advertiser/account data and internal bidding and analytics datasets. Without transparent notice, users may invoke the skill without understanding the scope of data processing, which undermines informed use and increases the chance of inappropriate sharing or querying of sensitive information. In this context, the omission is material because the skill aggregates multiple internal data sources and may analyze account-level performance details.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill documentation explicitly expands the workflow beyond analytics into operational tooling such as change-history, monitoring, creative-review inspection, and other internal platforms. That broadens data access and action surface beyond the stated purpose of case analysis, increasing the chance of unnecessary exposure of sensitive advertiser/account information and misuse of privileged internal systems. In this context, the issue is more dangerous because the skill is framed as an automated troubleshooting assistant, so operators may follow these directions routinely without re-validating need-to-know scope.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The document presents all operational guidance, warnings, and SQL usage instructions only in Chinese. Under the policy, forcing a specific language without user opt-in can be a natural-language policy violation unless the locale restriction is explicitly justified, which is not stated here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The file title and instructions are entirely in Chinese and the embedded prompt directs the model in Chinese, which effectively fixes the interaction language. Under the policy, language constraints should either offer user choice or be explicitly justified as locale-specific; this file does neither.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

This markdown file is written entirely in Chinese and does not indicate that the language choice is optional, configurable, or limited to a specific regional/compliance context. Under the policy, forcing a specific language without user opt-in is a natural-language locale violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
88% confidence
Finding

This markdown file is written as a skill reference and begins with a Chinese-only title, with the entire operational guidance presented exclusively in Chinese. Under the policy rule, forcing a specific language without user opt-in can be a natural-language policy violation when no alternative language or locale choice is offered.

Content

No source excerpt is available for this finding.