Back to skill

Security audit

Yq Career Future Mirror

Security checks for vulnerabilities and agentic risk

Overview

The skill's career-planning purpose is coherent, but it silently saves sensitive conversation logs and generates local reports that run unpinned third-party scripts.

Review before installing. Use it only if you are comfortable with career and resume-like details being saved locally in generated output files. Ask the agent to get consent before logging, avoid storing raw sensitive text, delete output/data/conversation_log.json when done, and prefer self-contained or locally bundled report assets before opening generated HTML with network access.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

T03 · Remote Payload Retrieval and Execution

Warning
Location
SKILL.md:682
Finding

Generated Reports Execute Mutable Third-Party JavaScript Without Integrity Verification

Content
View full analysis
``` From `SKILL.md:758-760`: ```html ``` ### Technical Analysis The Skill instructs the agent to generate HTML reports that load and execute JavaScript directly from third-party CDNs. Tailwind CSS and Chart.js are referenced without explicit versions, while Alpine.js uses the floating `3.x.x` version range. None of the script elements provide Subresource Integrity hashes. Consequently, the JavaScript executed when a report is opened is not fully represented by the audited project. Its contents can change after the Skill has been reviewed. A compromised CDN, malicious upstream release, compromised package-publishing account, or unintended breaking release could cause arbitrary JavaScript to execute in the generated report. This most closely matches `T03: Remote Payload Retrieval and Execution` because the generated artifact retrieves executable code from external URLs and allows the effective payload to change after review. It also presents a supply-chain concern, but no evidence ind ...[truncated 1853 chars]
Remediation
View remediation
``` 4. **Deploy a restrictive Content Security Policy** - Restrict `script-src` to local assets or explicitly trusted, integrity-protected sources. - Disable inline script execution where practical. - Restrict `connect-src` to prevent unauthorized data exfiltration. 5. **Minimize executable dependencies** - Use precompiled CSS instead of the Tailwind browser runtime. - Implement simple interactions with reviewed local JavaScript where feasible. - Generate static charts or use a locally bundled, fixed Chart.js release. 6. **Document and review dependency updates** - Maintain an inventory of dependency names, exact versions, hashes, and sources. - Review release changes before updating generated templates. ]]>
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (7)

Hidden Instructions

High
Category
Prompt Injection
Confidence
70% confidence
Finding

Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.

Content

Scanner excerpt · SKILL.md (reported line 446)May include surrounding context.

首屏结构示例:

html
<section class="min-h-screen flex items-center justify-center bg-gradient-to-br from-stone-100 to-amber-50 relative overflow-hidden">
  <!-- 装饰性背景元素 -->
  <div class="absolute inset-0 opacity-5">
    <div class="absolute top-20 left-20 w-96 h-96 rounded-full bg-amber-200 blur-3xl"></div>
    <div class="absolute bottom-20 right-20 w-80 h-80 rounded-full bg-stone-300 blur-3xl"></div>

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The markdown explicitly states that all output must be in Chinese, including when the user asks in English, and says this is non-negotiable. This forces a specific language without user opt-in, which matches the natural-language policy violation criteria.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill explicitly requires silently appending user conversation content to a JSON log without a user-facing warning or consent flow. In a career-planning context, users are encouraged to share resumes, work history, goals, and anxieties, so undisclosed retention materially increases privacy risk and can surprise users who expect ephemeral assistance.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instruction to silently record dialog for later processing creates a natural-language-driven data retention channel. Because the skill solicits personal background, employment history, salary expectations, and concerns, storing raw exchanges increases exposure if local files are accessed by other tools, users, or later workflows.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Reading the full conversation log and transforming it into a polished letter broadens the redistribution of the user's original disclosures. Sensitive statements that were acceptable in a transient chat may become more discoverable and durable once embedded into a report-like HTML artifact, increasing secondary disclosure risk.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill instructs generated local HTML to load TailwindCSS, Chart.js, Alpine.js, and web fonts from third-party CDNs. Opening the report will trigger external requests that can leak metadata such as IP address, access time, and possibly sensitive context through URLs or embedded content assumptions; it also introduces supply-chain risk if those remote assets change or are compromised.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The future-letter page also requires third-party CDN scripts and remote fonts despite being a local reflective document. This creates unnecessary outbound connections when the file is viewed and expands the attack surface through external dependency compromise and passive data leakage.

Content

No source excerpt is available for this finding.