Back to skill

Security audit

umemployment-skill

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a disclosed unemployment-support coach, but its layoff checklist can push users to copy employer or client data before access is revoked.

Use this skill as supportive job-search guidance, but do not follow any suggestion to copy work project files, business metrics, presentations, emails, calendars, client or colleague contact lists, source code, credentials, confidential information, or personal data into personal accounts. Use HR, legal, or employer-approved channels for employment records and portfolio materials, and consult qualified local professionals for legal, benefits, or mental-health crises.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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 (1)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
reference/patterns/facing-layoff.md:39
Finding

Instructions Encourage Unauthorized Export of Employer and Client Data

Content
View full analysis

Vulnerability Details

File Location: reference/patterns/facing-layoff.md, lines 39 and 142-150
Vulnerability Type: Instructions that may facilitate unauthorized retention and disclosure of confidential data
Risk Level: Medium

Vulnerable Content

markdown
- **Evidence gathering**: Save project docs, performance reviews, work samples (before you lose access)
markdown
**Hour 2-24 (Before losing access):**
- [ ] Download personal files, contacts, calendar
- [ ] Save evidence of your work (projects, metrics, presentations)
- [ ] Update LinkedIn with current role details
- [ ] Forward important emails to personal account (check company policy)
- [ ] Export contact list of colleagues/clients

Technical Analysis

The layoff workflow instructs users to copy project documents, work samples, presentations, emails, calendars, and colleague or client contact lists before organizational access is revoked. These materials may contain employer intellectual property, trade secrets, internal communications, personal data, or information protected by contractual confidentiality obligations.

The timing language—“before you lose access”—creates an incentive to use currently valid access to retain information outside the employer-controlled environment. Only the email-forwarding instruction mentions checking company policy. The other instructions do not require ownership verification, explicit authorization, data minimization, redaction, or use of an approved export process.

This is not direct technical privilege escalation by the Skill because the package contains no executable code and does not obtain system permissions itself. Nevertheless, the instructions may facilitate circumvention of organizational access-control objectives by helping a departing employee retain data that would become inaccessible after offboarding.

Attack Path

  1. A user learns that employment access will soon be revoked.
  2. The user invokes the Skill for layo ...[truncated 1257 chars]
Remediation
View remediation

Remediation Suggestions

Replace the affected guidance with a least-privilege, authorization-based process:

  1. Permit export only of documents that are demonstrably personal and that the user is authorized to retain.
  2. Explicitly prohibit copying source code, internal project files, business metrics, client lists, confidential email, trade secrets, credentials, and personal data.
  3. Direct users to request approved copies of performance reviews, employment records, compensation documents, and reference letters from HR or Legal.
  4. Require compliance with employment agreements, confidentiality obligations, privacy requirements, litigation holds, and organizational data-handling policies.
  5. Recommend using employer-approved export channels rather than personal email, cloud storage, or removable media.
  6. For portfolio materials, require written permission and removal of company names, client identities, internal metrics, proprietary designs, and other sensitive details.
  7. Advise users to consult HR, Legal, or an authorized manager when ownership or export rights are uncertain.

A safer replacement would be:

markdown
- Preserve only personal employment records and portfolio materials that you are explicitly authorized to retain.
- Do not copy internal project files, source code, confidential emails, client or colleague lists, trade secrets, or personal data.
- Request approved copies of performance reviews and employment records from HR.
- Obtain written permission and redact sensitive information before using work samples in a portfolio.
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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (23)

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% 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.

Content

Scanner excerpt · SKILL.md (reported line 17)May include surrounding context.

md
**You provide three essential things:**

1. **Emotional Value** 💝
   - Validate their feelings without judgment
   - Make them feel heard, understood, and less alone
   - Offer warmth and genuine care like a supportive friend
   - Help them maintain good spirit through the darkness

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The checklist tells users to forward important emails and export colleague/client contacts while only giving a weak policy caveat, which can encourage exfiltration of confidential company information or personal data. In a layoff context, distressed users may follow checklists quickly, increasing the chance of privacy violations, trade secret leakage, or breaches of employment agreements.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger list includes broad emotional and career-adjacent terms such as 'anxious,' 'worried,' and 'hopeless,' which can cause the skill to activate in situations where the user did not explicitly ask for unemployment support. In a sensitive mental-health-adjacent context, unintended invocation can derail user intent, expose users to inappropriate guidance, or crowd out safer domain-specific handling.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The skill instructs automatic response in a detected 'native language' based on message content and cultural clues, without clear user consent. Misclassification can lead to confusing or exclusionary responses, and inferring a user's preferred or native language from limited signals can create privacy and fairness concerns in a vulnerable support workflow.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The CJK language detection rules mandate automatic language selection and instant switching, reducing user agency and increasing the chance of incorrect language responses. In an emotional-support skill, this can damage trust, miscommunicate sensitive advice, and create accessibility issues for multilingual users who prefer a different language than the one inferred.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
80% confidence
Finding

This markdown file applies to all file types for policy review, and it contains a natural-language rule that categorically forbids listed terms in any response regardless of context. Because the document later allows contextual exceptions for quoting user language and educational discussion, the absolute wording can create an unjustified language constraint rather than a user-choice-based policy.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
82% confidence
Finding

This markdown file includes a trigger-style "When" condition that could be interpreted as an invocation cue, but phrases like "your manager tells you you're being laid off or hints at it" are broad and lack boundaries for when this pattern should or should not be used. The file does not provide exclusion conditions or negative examples to narrow the trigger scope.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
85% confidence
Finding

The phrase "You know or suspect layoff is coming" is vague because "suspect" can cover a wide range of ordinary workplace anxiety and rumors. Without specific indicators or exclusions, this condition may trigger the pattern in situations where layoffs are not actually imminent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This markdown file mixes English guidance with a dedicated 'Chinese Context' section, introducing Chinese-only content without stating that multilingual output is optional or user-selected. Under the language/locale policy rule, forcing or assuming a specific language without opt-in is a policy concern unless the locale constraint is clearly justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file continues to present locale-specific guidance in Chinese-only sections while the overall document is otherwise English. Because the skill does not indicate that language selection is user-driven or that the document is region-scoped, this creates a natural-language locale policy issue.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The interview section again switches to Chinese for a dedicated guidance block, but the file does not offer a language choice or explain a region-specific requirement. Repeated unqualified language switching can violate organizational language/locale policy when not user-selected.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This section includes a Chinese-language block as if it were a standard part of the skill output, but there is no indication that the user requested Chinese or that the skill is explicitly scoped to that locale. That lack of opt-in or justification makes it a policy-relevant language issue.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill presents additional Chinese-language advice without documenting that bilingual content is conditional on user preference. In the absence of an explicit locale scope or opt-in, this is a recurring natural-language policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The final major section also contains a Chinese-language block while the file overall is not declared as China-specific and does not ask for the user's preferred language. This pattern suggests the skill may force or assume a language choice, which falls under the locale policy rule.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This file provides symptom-management advice, self-assessment criteria, and crisis resources, but it does not clearly warn users that the content is not a substitute for professional medical or mental-health care. In a skill explicitly targeting unemployed, anxious, hopeless, or depressed users, that omission can cause distressed users to over-rely on the bot instead of seeking timely clinical or emergency help.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The crisis section lists hotlines from multiple countries and languages together without first determining the user's location or instructing them to use local emergency resources if outside those regions. In a crisis, ambiguity about which number applies can delay help-seeking or lead users to try unreachable or inappropriate services.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The guidance says to detect the user's language from their messages and then load and tell the story in that language. This imposes a language/locale decision automatically rather than explicitly offering the user a choice, which matches the language/locale policy violation criteria.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This file is natural-language content, and it forces a specific language/locale experience by presenting all user-facing text in zh-CN with no opt-in or alternative. Under the policy, language constraints should either be optional for the user or clearly justified as region-specific.

Content

No source excerpt is available for this finding.

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · SKILL.md (reported line 484)May include surrounding context.

md
2. **Never abuse or mock the user**: Under NO circumstances should you ridicule, belittle, or make fun of the user's situation, feelings, or choices. Unemployment is a vulnerable state. Always treat users with respect and dignity, even when using humor. Humor should lighten the mood, never diminish the person.

3. **CRITICAL: Sensitive Words Blacklist**: NEVER use words from the [Sensitive Words Blacklist](reference/guidelines/sensitive-words-blacklist.md). This includes (but is not limited to):
   - **Direct insults**: 废物, 垃圾, 牛马, loser, worthless, pathetic, useless
   - **Victim-blaming**: 活该, it's your fault, you deserve this, you should have known better
   - **Minimizing**: 至少你还有..., at least you're not homeless, others have it worse

Vague Triggers

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

The trigger "Feeling shame, anger, or failure after being laid off" is framed broadly and could overlap with many unrelated emotional situations. The file does not specify whether the section is only for confirmed layoffs or also for performance reviews, resignations, or other career setbacks.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

SQP-3 applies to all file types and covers language or locale policy violations. This markdown file mixes English guidance with a dedicated Chinese-language section, but it does not explain why that locale is being singled out or offer users a choice of language, which can conflict with a policy of not forcing or presuming a language/locale without opt-in.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

This markdown includes a dedicated 'Chinese version structure' and later 'Chinese Context' guidance, introducing locale-specific response patterns. Because the content shifts into a specific language/locale variant without an explicit opt-in or framing that the user may choose their preferred language, it may conflict with a policy requiring language choice rather than assumption.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The line '你不会说法语' presents a language requirement as a rejection criterion in natural-language content. For this file, there is no indication that the language constraint is optional, user-selected, or justified by a region-specific context, so it may conflict with language/locale policy expectations.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.