Security audit

therapist-screener（甄选心理咨询师）

Security checks for vulnerabilities and agentic risk

Overview

This is a plain Markdown therapist-screening guide with no executable code, persistence, credential handling, or hidden behavior, though users should confirm the relevant jurisdiction instead of relying on language defaults.

Before relying on an evaluation, tell the agent the therapist's practice location or licensing jurisdiction, especially if the conversation language differs from where the therapist works. Prefer pasting profile text over sending links when possible, and treat the report as screening support rather than medical, legal, or licensing advice.

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 (3)

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The skill description contains a very broad set of trigger terms covering generic requests like finding, evaluating, or choosing a therapist in both Chinese and English. This can cause the skill to activate for wide classes of mental-health-related user queries without clear user intent, increasing the chance of unrequested routing into a prescriptive workflow in a sensitive domain.

Natural-Language Policy Violations

Medium

Confidence: 89% confidence
Finding: The skill automatically routes users to a domestic or overseas evaluation mode based solely on conversation language. In a regulated and culturally sensitive domain like therapist credential evaluation, language is an unreliable proxy for jurisdiction, so users may receive the wrong licensing, certification, and standards framework without realizing it.

Natural-Language Policy Violations

Medium

Confidence: 95% confidence
Finding: The language detection section explicitly mandates locale-specific behavior based on whether the conversation is in Chinese or English unless the user overrides it. This creates silent policy switching in a mental-health screening context, which can misapply credential standards and produce materially misleading guidance about therapist qualifications or red flags.

Static analysis

No suspicious patterns detected.