Back to skill

Security audit

Dealpilot

Security checks for vulnerabilities and agentic risk

Overview

DealPilot is a disclosed prototype that returns mock shopping advice, with no evidence of malicious access or persistence, but its recommendations should not be treated as live pricing.

Install only if you understand this is a prototype skeleton. Do not rely on its generated prices, subsidy claims, seller-risk claims, or buy-now advice for actual purchases until live platform adapters and clear mock-data warnings are added.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T09 · Insecure Skill Coding Practices

Warning
Location
engine/router.ts:16
Finding

Unverified Mock Data Presented as Actionable Shopping Advice

Content
View full analysis

Vulnerability Details

File Location: engine/router.ts, lines 16-63
Vulnerability Type: Unverified fixed data presented as current pricing and purchasing intelligence
Risk Level: Medium

Code Evidence

typescript
export async function runDecisionEngine(
  request: ShoppingRequest
): Promise<DecisionReport> {
  const platforms: Platform[] = request.platforms || ["taobao", "pdd", "jd", "yhd", "vip"];

  return buildMockReport(request, platforms);
}

function buildMockReport(request: ShoppingRequest, platforms: Platform[]): DecisionReport {
  const product = request.product || "Unknown product";

  return {
    recommendedPlatform: "pdd",
    recommendedPrice: 89,
typescript
timingAdvice: {
  verdict: "buy_now",
},
comparison: platforms.map((p) => ({
  platform: p,
  price: p === "pdd" ? 89 : p === "jd" ? 109 : p === "taobao" ? 95 : p === "yhd" ? 92 : 88,
  quality: p === "jd" ? "high" : p === "taobao" ? "medium" : "medium",
  shipping: p === "jd" ? "fast" : p === "yhd" ? "fast" : p === "pdd" ? "slow" : "medium",
  afterSales: p === "jd" ? "strong" : p === "taobao" ? "medium" : "weak",
  authenticity: p === "jd" ? "safe" : p === "pdd" ? "risky" : "medium",
  score: p === "pdd" ? 85 : p === "jd" ? 82 : p === "taobao" ? 75 : 70,
})),

The resulting data is formatted as an ordinary recommendation without a mock-data warning:

javascript
export function formatReport(report) {
  let out = "";
  if (report.recommendedPrice) out += `Reference price: ¥${report.recommendedPrice}\n`;

  report.reasons.forEach((r) => (out += `- ${r}\n`));

  out += `${report.timingAdvice.reason}\n`;

  report.alternatives.forEach((a) => {
    out += `- **${a.platform}** (${a.score}) ${a.price ? `¥${a.price}` : ""}: ${a.keyReason}\n`;
  });

  return out;
}

Technical Analysis

The decision engine does not query any platform adapter, retrieve current prices, inspect sellers, or obtain historical pricing data. Instead, every request ...[truncated 2207 chars]

Remediation
View remediation

Remediation Suggestions

  1. Prevent the Skill from issuing actionable purchase recommendations until verified platform data is available.
  2. Add a prominent runtime warning to every stub report stating that all prices, scores, promotions, seller assessments, and timing advice are synthetic test data.
  3. Remove claims about current prices, official subsidies, historical lows, authenticity, and seller protections unless they are supported by retrieved evidence.
  4. Query actual platform adapters and include the source platform, retrieval timestamp, seller identity, and product URL for every recommendation.
  5. Enforce the caller's platform allowlist. The recommended platform must be selected only from request.platforms.
  6. Validate and apply budget, quantity, priorities, urgency, scenario, mustHave, and mustAvoid constraints before constructing a recommendation.
  7. Return compare_first or an explicit insufficient-data result when live data cannot be obtained.
  8. Separate mock and production modes through an explicit configuration flag, and prevent production deployments from enabling mock recommendations.
  9. Add tests confirming that:
    • Excluded platforms cannot be recommended.
    • Prices are not emitted without a verified source.
    • Stub output always contains a clear warning.
    • Unsupported historical-price claims are rejected.
    • User constraints materially affect the result.
Vulnerability Patterns
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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 (13)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 250)May include surrounding context.

md
const { formatReport } = require('./scripts/analyze.js');

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

This markdown file contains user-facing operational instructions only in Chinese, including headings, steps, troubleshooting, and examples. Under the policy rule, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is explicitly justified, which is not present here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The heading and core instructions are written as direct Chinese-language role guidance, including '你是全网购物决策官' and the task description, but there is no indication that users may choose another language. This creates a natural-language locale policy issue because the skill appears to enforce a specific language without user opt-in or documented justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The description is written entirely in Chinese and presents the skill in that language without any indication of user language selection or an explicit region-specific limitation. This can violate language/locale policy when a skill implicitly requires a specific language without opt-in or justification.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The file comments claim the router receives requests, dispatches to platform adapters, and aggregates results, but the implementation does none of that and instead returns synthesized output. This documentation/behavior mismatch is security-relevant because users and downstream systems may rely on the skill for risk-aware shopping decisions while the reported comparison is entirely unverified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This TypeScript file contains comments and returned recommendation strings entirely in Chinese, including the final conclusion shown to users. The skill does not indicate any user-selectable language or region-specific justification, which can violate a language/locale policy requiring opt-in rather than forcing a specific language.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill advertises cross-platform shopping comparison, but the router returns a fixed mock recommendation instead of using live or adapter-derived platform data. In a purchasing-decision agent, this can mislead users into making financial decisions based on fabricated results, undermining trust and potentially steering them toward unsafe or suboptimal sellers.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This JavaScript file formats all user-facing report text in Chinese, including headings, recommendations, and conclusions. The skill does not offer a language option or document a justified region-specific constraint, which creates a natural-language locale policy violation under the stated rules.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The README title and core description use Chinese alongside English, which can impose a language assumption on users without explicit opt-in or documented locale scope. Under the policy rule, forced or assumed language usage should either offer user choice or clearly justify the locale constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The comment DealPilot 类型定义 is a natural-language indicator that the skill content is localized to Chinese, but this file provides no user choice or documented opt-in for language/locale. Under the policy rule, forcing a specific language without an explicit choice can be a locale-policy issue.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file contains natural-language comments and thrown error strings exclusively in Chinese, which imposes a specific language choice on maintainers or users interacting with these messages. No opt-in, alternative locale, or documented regional justification is present in the file.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

This code uses Chinese-only natural-language comments and sets the default priorities to Chinese strings ("价格", "品质"). That can impose a language/locale choice in the skill behavior without offering the user any opt-in or alternative, which matches the policy's language/locale violation criterion.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

This JavaScript file contains multiple user-facing console messages in Chinese, including the script title and status output. Under the policy, forcing a specific language without user choice is a natural-language policy violation unless the locale constraint is clearly documented and justified, which is not evident in this file.

Content

No source excerpt is available for this finding.