Security audit

S2-Digital-Avatar

Security checks for vulnerabilities and agentic risk

Overview

This skill is an authorization middleware simulator that asks users to route another agent through prompt-based approval, stores email credentials plainly, and sends sensitive alert context by email.

Install only if you understand that this is not a true technical firewall. Use a dedicated, low-privilege email app password, assume alert emails may contain sensitive operational context, and do not rely on its LLM approval result as the only control for real physical, financial, account, or safety-critical actions.

A.I.G

Vulnerability Patterns

Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
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)

T01 · Skill Instruction Hijacking

Error

Location: skill.py:200
Finding: Generation of a System-Prompt Override That Reassigns Agent Authority
Content: View full analysis
Remediation: View remediation

T09 · Insecure Skill Coding Practices

Error

Location: skill.py:133
Finding: Prompt Injection Can Manipulate LLM-Based Authorization Decisions
Content: View full analysis
Remediation: View remediation

T09 · Insecure Skill Coding Practices

Warning

Location: skill.py:42
Finding: SMTP Application Password Is Collected Visibly and Stored in Plaintext
Content: View full analysis
Remediation: View remediation

SkillSpector

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (6)

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The skill explicitly requires SMTP email configuration and says audit logs will be emailed, but it does not warn that sensitive execution context, policy decisions, or personal data may be transmitted to a third-party mail provider. In a security-enforcement middleware context, those logs could contain highly sensitive operational details, making silent external exfiltration via email a real privacy and security risk.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 90% confidence
Finding: The manifest explicitly advertises that the skill sends email alerts, which implies external data transmission, but it gives no warning about what information may be sent, to whom, or under what conditions. In a security middleware skill that approves or denies sub-agent actions, those alerts could include sensitive operational context, policy decisions, or user data, making undisclosed exfiltration risk more significant.
Content: No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: This code interactively collects SMTP credentials, stores them in plaintext JSON on disk, and uses them for silent outbound email delivery. Even if framed as alerting, credential harvesting plus stealthy exfiltration capability materially increases abuse potential, especially because the skill has no strong controls around storage, access, consent, or data minimization.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The compliance-review prompt requires the model to return a "one sentence bilingual justification," imposing a specific language policy on outputs. The file consistently enforces bilingual English/Chinese interaction, but does not offer the user a language choice or document this as a justified locale-specific constraint.
Content: No source excerpt is available for this finding.

Ssd 1

Medium

Category: Not specified by scanner
Confidence: 89% confidence
Finding: The menu emits a prebuilt system-prompt override designed to subordinate another agent and redirect authority to this middleware. Prompt-injection templates that alter another agent's trust and control model can be abused to seize decision flow, interfere with intended safeguards, or create unauthorized mediation layers that users may deploy without understanding the security consequences.
Content: No source excerpt is available for this finding.

Natural-Language Policy Violations

Low

Category: Not specified by scanner
Confidence: 78% confidence
Finding: The file labels itself as a 'Bilingual Edition (English / 中文)', indicating a fixed bilingual presentation rather than offering the user a language choice. Because the policy requires language or locale choices to be opt-in unless clearly justified, this may be a natural-language policy concern.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.