Security audit

qingtian-free-psych-resources

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches its mental-health resource and self-assessment purpose, but it needs review because several high-risk questionnaire items are not connected to crisis handling and sensitive reports may be saved locally.

Review before installing if you may use the assessment module with distressed users. Treat it as informational only, confirm the user's location before relying on crisis resources, avoid saving reports in shared or synced folders unless the user explicitly agrees, and do not rely on SDS/CES-D/SCL-90 or FFMQ outputs for urgent safety decisions until the crisis metadata and scoring keys are corrected.

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
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (71)

Missing User Warnings

High

Confidence: 97% confidence
Finding: This self-assessment contains severe depressive content but provides no warning, no crisis disclaimer, and no crisis-response guidance in the interpretation or safety metadata. If a distressed user endorses high-risk statements, the system may respond with only routine scoring feedback, delaying appropriate intervention or help-seeking.

Intent-Code Divergence

High

Confidence: 99% confidence
Finding: All '不反应' items (nr1-nr7) are written as positively keyed abilities such as tolerating distress without reacting, yet the metadata marks them reverse-scored. This would systematically invert the entire nonreactivity factor, causing users with stronger mindfulness traits to receive worse scores and potentially misleading downstream interpretations or recommendations.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The self-harm item '想结束自己的生命' is present but has crisis:false and no associated warning or escalation metadata. That omission is dangerous because a user can disclose suicidal intent without the system recognizing it as an emergency signal requiring immediate intervention.

Intent-Code Divergence

High

Confidence: 99% confidence
Finding: The file contains explicit self-harm and violence-related items, such as suicidal ideation and harming others, yet the safety metadata declares no crisis items and provides no escalation rules. In a mental-health assessment skill, this can cause downstream systems to miss urgent risk states and fail to trigger warnings, crisis guidance, or human escalation.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The global safety object is effectively empty despite this questionnaire containing multiple high-risk mental-health prompts, including suicide, death ideation, psychotic symptoms, and violence-related items. An empty safety section in this context creates a systemic safety failure because consuming applications may rely on it to determine crisis workflows.

Intent-Code Divergence

High

Confidence: 98% confidence
Finding: This scale includes a suicidality-related item (q19: '假若我死了别人会过得更好') yet the safety block declares no crisis items and provides no crisis or high-risk rule. In a mental-health assessment context, that mismatch can cause the surrounding system to miss or suppress escalation for users expressing suicidal ideation, creating a significant risk of harmful non-response.

Missing User Warnings

High

Confidence: 97% confidence
Finding: The file presents a depression self-assessment containing a suicidality-related question but gives no warning, no crisis disclaimer, and no escalation guidance for positive responses. In practice, this can lead users with suicidal thoughts to receive only routine scoring output instead of urgent safety guidance, which is especially dangerous in self-serve mental-health tools.

Vague Triggers

Medium

Confidence: 96% confidence
Finding: The trigger list includes broad, everyday phrases such as '想找人说说话' and generic help/menu terms, which can cause the skill to activate in situations where the user did not specifically request this resource finder. In a mental-health context, unintended invocation is more sensitive because users may be discussing distress, crisis, or private issues, and the skill could inappropriately steer the interaction toward its own workflow or resource set.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The README states that generated psychological assessment reports are saved by default to the current working directory, but it does not warn users that these reports may contain highly sensitive mental-health information. In shared machines, synced folders, CI workspaces, or project directories with broad access, this default can expose private data to unauthorized users or persistence mechanisms.

Natural-Language Policy Violations

Medium

Confidence: 93% confidence
Finding: The skill summary and description indicate Chinese report generation as the default behavior, with no explicit user language selection or opt-in. This is not inherently dangerous by itself, but it can create comprehension and consent issues if a user does not read Chinese well, especially for sensitive mental-health results, safety notices, or crisis guidance.

Vague Triggers

Medium

Confidence: 96% confidence
Finding: The activation words include extremely common terms like “启动 / 开始 / start”, which can appear in ordinary conversation and may invoke the skill without clear user intent. In a mental-health assessment context, accidental activation is more sensitive because it can cause the agent to begin collecting psychological self-assessment data or steer the conversation into a clinical-style workflow unexpectedly.

Vague Triggers

Medium

Confidence: 88% confidence
Finding: The trigger scenarios are broad natural-language intents such as wanting to measure anxiety, depression, stress, or asking for a report, without clear exclusion boundaries. This can cause over-triggering on general wellness or emotional-support conversations and prematurely route users into a structured assessment flow, which is particularly risky in a sensitive mental-health domain where context and consent matter.

Natural-Language Policy Violations

Medium

Confidence: 89% confidence
Finding: All user-facing instructions and examples in this file are presented in Chinese, and there is no indication that the user can choose another language or that the skill is intentionally restricted to a Chinese-speaking audience. This can violate language or locale policy when a specific language is effectively forced without opt-in.

Vague Triggers

Medium

Confidence: 94% confidence
Finding: The activation example uses the generic phrase “启动”, which can easily occur in normal conversation and may unintentionally trigger the skill. In a mental-health assessment context, accidental activation is more concerning because it can start a sensitive workflow, prompt disclosure of personal information, or generate reports without clear user intent.

Natural-Language Policy Violations

Medium

Confidence: 93% confidence
Finding: The protocol hard-codes only China-specific crisis resources and emergency numbers without first confirming the user's location or offering jurisdiction-appropriate alternatives. In a mental-health crisis context, giving regionally incorrect hotline or emergency guidance can delay access to help, which materially increases safety risk even though the document's intent is protective.

Natural-Language Policy Violations

Medium

Confidence: 94% confidence
Finding: The document declares `lang="zh-CN"`, and all user-facing content is presented only in Simplified Chinese. Under the policy, forcing a specific language without user opt-in is a natural-language policy concern unless the locale restriction is explicitly justified, which is not present in this file.

Natural-Language Policy Violations

Medium

Confidence: 95% confidence
Finding: The HTML declares `lang="zh-CN"` and the title/content are presented entirely in Chinese, indicating a fixed locale. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale limitation is explicitly justified, which is not present in this file.

Intent-Code Divergence

Medium

Confidence: 98% confidence
Finding: The report shows a standard score of 62.5 with a mild depression classification, but the escalation guidance states that a standard score of 53 already enters the mild depression range, creating inconsistent clinical messaging. In a mental-health assessment context, contradictory thresholds can mislead users about severity and delay appropriate care-seeking or distort risk perception, which makes this materially safety-relevant.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: This JSON defines a mental-health self-assessment with score bands, summaries, and behavioral advice, but it does not include any user-facing limitation notice that the result is only a screening or self-reflection aid and not a clinical diagnosis. In a mental-health context, users may over-trust the output, self-misclassify their resilience, or substitute the tool for professional evaluation, especially because the file also contains an effectively empty safety section.

Natural-Language Policy Violations

Medium

Confidence: 83% confidence
Finding: User-facing fields including the scale name, description, questions, and interpretation are entirely in Chinese, and the file does not indicate that the skill is China/Chinese-only or offer any user language selection. Under the stated policy, forcing a specific language without opt-in or clear locale justification is a natural-language policy concern.

Intent-Code Divergence

Medium

Confidence: 96% confidence
Finding: The scale includes an item stating '我无法继续我的生活' ('I cannot go on with my life'), which is crisis-indicative, yet the safety section leaves crisis_items, crisis_rule, and high_risk_rule empty and all items are marked crisis=false. In a mental-health screening context, this mismatch can suppress escalation logic and fail to trigger urgent guidance for potentially suicidal or severely impaired users.

Missing User Warnings

Medium

Confidence: 94% confidence
Finding: This file provides mental-health style score bands and advice such as suggesting counseling, but it does not include any limitation notice that the scale is informational only and not a diagnosis. In the context of a self-administered attachment assessment, users may over-trust the output, self-label, or make relationship or treatment decisions without professional evaluation.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The questionnaire contains emotionally sensitive items about abandonment, lovability, jealousy, and fear of closeness, yet the safety block leaves crisis_items, crisis_rule, and high_risk_rule empty. Without distress guidance or escalation logic, a vulnerable user could be left unsupported if the content triggers anxiety, relationship distress, or broader emotional crisis.

Natural-Language Policy Violations

Medium

Confidence: 93% confidence
Finding: This JSON manifest contains all user-facing strings in Chinese, including the scale name, description, items, interpretations, and advice, but provides no indication that the skill is China-specific or that users can opt into this locale. Under the language/locale policy, hard-coding a single language without opt-in or justification is a natural-language policy concern.

Intent-Code Divergence

Medium

Confidence: 97% confidence
Finding: Item ob2 states 'I do not notice many of the emotions and sensations I am experiencing,' which is negatively keyed content, but the metadata marks it as non-reverse. This will invert the meaning of responses for that item and distort the '观察' dimension score, producing inaccurate psychological assessment output.

Static analysis

No suspicious patterns detected.